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_256_single kernel generator.
44 #include "../nb_kernel.h"
45 #include "gromacs/legacyheaders/types/simple.h"
46 #include "gromacs/math/vec.h"
47 #include "gromacs/legacyheaders/nrnb.h"
49 #include "gromacs/simd/math_x86_avx_256_single.h"
50 #include "kernelutil_x86_avx_256_single.h"
53 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwNone_GeomW4P1_VF_avx_256_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_256_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,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight 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 jnrE,jnrF,jnrG,jnrH;
78 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
79 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
80 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
81 int j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
82 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
84 real *shiftvec,*fshift,*x,*f;
85 real *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
87 __m256 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
88 real * vdwioffsetptr1;
89 __m256 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
90 real * vdwioffsetptr2;
91 __m256 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
92 real * vdwioffsetptr3;
93 __m256 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
94 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
95 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
96 __m256 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
97 __m256 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
98 __m256 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
99 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
102 __m128i vfitab_lo,vfitab_hi;
103 __m128i ifour = _mm_set1_epi32(4);
104 __m256 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
106 __m256 dummy_mask,cutoff_mask;
107 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
108 __m256 one = _mm256_set1_ps(1.0);
109 __m256 two = _mm256_set1_ps(2.0);
115 jindex = nlist->jindex;
117 shiftidx = nlist->shift;
119 shiftvec = fr->shift_vec[0];
120 fshift = fr->fshift[0];
121 facel = _mm256_set1_ps(fr->epsfac);
122 charge = mdatoms->chargeA;
124 vftab = kernel_data->table_elec->data;
125 vftabscale = _mm256_set1_ps(kernel_data->table_elec->scale);
127 /* Setup water-specific parameters */
128 inr = nlist->iinr[0];
129 iq1 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
130 iq2 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
131 iq3 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
133 /* Avoid stupid compiler warnings */
134 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
147 for(iidx=0;iidx<4*DIM;iidx++)
152 /* Start outer loop over neighborlists */
153 for(iidx=0; iidx<nri; iidx++)
155 /* Load shift vector for this list */
156 i_shift_offset = DIM*shiftidx[iidx];
158 /* Load limits for loop over neighbors */
159 j_index_start = jindex[iidx];
160 j_index_end = jindex[iidx+1];
162 /* Get outer coordinate index */
164 i_coord_offset = DIM*inr;
166 /* Load i particle coords and add shift vector */
167 gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
168 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
170 fix1 = _mm256_setzero_ps();
171 fiy1 = _mm256_setzero_ps();
172 fiz1 = _mm256_setzero_ps();
173 fix2 = _mm256_setzero_ps();
174 fiy2 = _mm256_setzero_ps();
175 fiz2 = _mm256_setzero_ps();
176 fix3 = _mm256_setzero_ps();
177 fiy3 = _mm256_setzero_ps();
178 fiz3 = _mm256_setzero_ps();
180 /* Reset potential sums */
181 velecsum = _mm256_setzero_ps();
183 /* Start inner kernel loop */
184 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
187 /* Get j neighbor index, and coordinate index */
196 j_coord_offsetA = DIM*jnrA;
197 j_coord_offsetB = DIM*jnrB;
198 j_coord_offsetC = DIM*jnrC;
199 j_coord_offsetD = DIM*jnrD;
200 j_coord_offsetE = DIM*jnrE;
201 j_coord_offsetF = DIM*jnrF;
202 j_coord_offsetG = DIM*jnrG;
203 j_coord_offsetH = DIM*jnrH;
205 /* load j atom coordinates */
206 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
207 x+j_coord_offsetC,x+j_coord_offsetD,
208 x+j_coord_offsetE,x+j_coord_offsetF,
209 x+j_coord_offsetG,x+j_coord_offsetH,
212 /* Calculate displacement vector */
213 dx10 = _mm256_sub_ps(ix1,jx0);
214 dy10 = _mm256_sub_ps(iy1,jy0);
215 dz10 = _mm256_sub_ps(iz1,jz0);
216 dx20 = _mm256_sub_ps(ix2,jx0);
217 dy20 = _mm256_sub_ps(iy2,jy0);
218 dz20 = _mm256_sub_ps(iz2,jz0);
219 dx30 = _mm256_sub_ps(ix3,jx0);
220 dy30 = _mm256_sub_ps(iy3,jy0);
221 dz30 = _mm256_sub_ps(iz3,jz0);
223 /* Calculate squared distance and things based on it */
224 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
225 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
226 rsq30 = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
228 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
229 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
230 rinv30 = gmx_mm256_invsqrt_ps(rsq30);
232 /* Load parameters for j particles */
233 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
234 charge+jnrC+0,charge+jnrD+0,
235 charge+jnrE+0,charge+jnrF+0,
236 charge+jnrG+0,charge+jnrH+0);
238 fjx0 = _mm256_setzero_ps();
239 fjy0 = _mm256_setzero_ps();
240 fjz0 = _mm256_setzero_ps();
242 /**************************
243 * CALCULATE INTERACTIONS *
244 **************************/
246 r10 = _mm256_mul_ps(rsq10,rinv10);
248 /* Compute parameters for interactions between i and j atoms */
249 qq10 = _mm256_mul_ps(iq1,jq0);
251 /* Calculate table index by multiplying r with table scale and truncate to integer */
252 rt = _mm256_mul_ps(r10,vftabscale);
253 vfitab = _mm256_cvttps_epi32(rt);
254 vfeps = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
255 /* AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
256 vfitab_lo = _mm256_extractf128_si256(vfitab,0x0);
257 vfitab_hi = _mm256_extractf128_si256(vfitab,0x1);
258 vfitab_lo = _mm_slli_epi32(vfitab_lo,2);
259 vfitab_hi = _mm_slli_epi32(vfitab_hi,2);
261 /* CUBIC SPLINE TABLE ELECTROSTATICS */
262 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
263 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
264 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
265 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
266 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
267 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
268 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
269 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
270 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
271 Heps = _mm256_mul_ps(vfeps,H);
272 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
273 VV = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
274 velec = _mm256_mul_ps(qq10,VV);
275 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
276 felec = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
278 /* Update potential sum for this i atom from the interaction with this j atom. */
279 velecsum = _mm256_add_ps(velecsum,velec);
283 /* Calculate temporary vectorial force */
284 tx = _mm256_mul_ps(fscal,dx10);
285 ty = _mm256_mul_ps(fscal,dy10);
286 tz = _mm256_mul_ps(fscal,dz10);
288 /* Update vectorial force */
289 fix1 = _mm256_add_ps(fix1,tx);
290 fiy1 = _mm256_add_ps(fiy1,ty);
291 fiz1 = _mm256_add_ps(fiz1,tz);
293 fjx0 = _mm256_add_ps(fjx0,tx);
294 fjy0 = _mm256_add_ps(fjy0,ty);
295 fjz0 = _mm256_add_ps(fjz0,tz);
297 /**************************
298 * CALCULATE INTERACTIONS *
299 **************************/
301 r20 = _mm256_mul_ps(rsq20,rinv20);
303 /* Compute parameters for interactions between i and j atoms */
304 qq20 = _mm256_mul_ps(iq2,jq0);
306 /* Calculate table index by multiplying r with table scale and truncate to integer */
307 rt = _mm256_mul_ps(r20,vftabscale);
308 vfitab = _mm256_cvttps_epi32(rt);
309 vfeps = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
310 /* AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
311 vfitab_lo = _mm256_extractf128_si256(vfitab,0x0);
312 vfitab_hi = _mm256_extractf128_si256(vfitab,0x1);
313 vfitab_lo = _mm_slli_epi32(vfitab_lo,2);
314 vfitab_hi = _mm_slli_epi32(vfitab_hi,2);
316 /* CUBIC SPLINE TABLE ELECTROSTATICS */
317 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
318 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
319 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
320 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
321 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
322 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
323 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
324 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
325 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
326 Heps = _mm256_mul_ps(vfeps,H);
327 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
328 VV = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
329 velec = _mm256_mul_ps(qq20,VV);
330 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
331 felec = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
333 /* Update potential sum for this i atom from the interaction with this j atom. */
334 velecsum = _mm256_add_ps(velecsum,velec);
338 /* Calculate temporary vectorial force */
339 tx = _mm256_mul_ps(fscal,dx20);
340 ty = _mm256_mul_ps(fscal,dy20);
341 tz = _mm256_mul_ps(fscal,dz20);
343 /* Update vectorial force */
344 fix2 = _mm256_add_ps(fix2,tx);
345 fiy2 = _mm256_add_ps(fiy2,ty);
346 fiz2 = _mm256_add_ps(fiz2,tz);
348 fjx0 = _mm256_add_ps(fjx0,tx);
349 fjy0 = _mm256_add_ps(fjy0,ty);
350 fjz0 = _mm256_add_ps(fjz0,tz);
352 /**************************
353 * CALCULATE INTERACTIONS *
354 **************************/
356 r30 = _mm256_mul_ps(rsq30,rinv30);
358 /* Compute parameters for interactions between i and j atoms */
359 qq30 = _mm256_mul_ps(iq3,jq0);
361 /* Calculate table index by multiplying r with table scale and truncate to integer */
362 rt = _mm256_mul_ps(r30,vftabscale);
363 vfitab = _mm256_cvttps_epi32(rt);
364 vfeps = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
365 /* AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
366 vfitab_lo = _mm256_extractf128_si256(vfitab,0x0);
367 vfitab_hi = _mm256_extractf128_si256(vfitab,0x1);
368 vfitab_lo = _mm_slli_epi32(vfitab_lo,2);
369 vfitab_hi = _mm_slli_epi32(vfitab_hi,2);
371 /* CUBIC SPLINE TABLE ELECTROSTATICS */
372 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
373 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
374 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
375 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
376 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
377 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
378 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
379 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
380 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
381 Heps = _mm256_mul_ps(vfeps,H);
382 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
383 VV = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
384 velec = _mm256_mul_ps(qq30,VV);
385 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
386 felec = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq30,FF),_mm256_mul_ps(vftabscale,rinv30)));
388 /* Update potential sum for this i atom from the interaction with this j atom. */
389 velecsum = _mm256_add_ps(velecsum,velec);
393 /* Calculate temporary vectorial force */
394 tx = _mm256_mul_ps(fscal,dx30);
395 ty = _mm256_mul_ps(fscal,dy30);
396 tz = _mm256_mul_ps(fscal,dz30);
398 /* Update vectorial force */
399 fix3 = _mm256_add_ps(fix3,tx);
400 fiy3 = _mm256_add_ps(fiy3,ty);
401 fiz3 = _mm256_add_ps(fiz3,tz);
403 fjx0 = _mm256_add_ps(fjx0,tx);
404 fjy0 = _mm256_add_ps(fjy0,ty);
405 fjz0 = _mm256_add_ps(fjz0,tz);
407 fjptrA = f+j_coord_offsetA;
408 fjptrB = f+j_coord_offsetB;
409 fjptrC = f+j_coord_offsetC;
410 fjptrD = f+j_coord_offsetD;
411 fjptrE = f+j_coord_offsetE;
412 fjptrF = f+j_coord_offsetF;
413 fjptrG = f+j_coord_offsetG;
414 fjptrH = f+j_coord_offsetH;
416 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
418 /* Inner loop uses 132 flops */
424 /* Get j neighbor index, and coordinate index */
425 jnrlistA = jjnr[jidx];
426 jnrlistB = jjnr[jidx+1];
427 jnrlistC = jjnr[jidx+2];
428 jnrlistD = jjnr[jidx+3];
429 jnrlistE = jjnr[jidx+4];
430 jnrlistF = jjnr[jidx+5];
431 jnrlistG = jjnr[jidx+6];
432 jnrlistH = jjnr[jidx+7];
433 /* Sign of each element will be negative for non-real atoms.
434 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
435 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
437 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
438 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
440 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
441 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
442 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
443 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
444 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
445 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
446 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
447 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
448 j_coord_offsetA = DIM*jnrA;
449 j_coord_offsetB = DIM*jnrB;
450 j_coord_offsetC = DIM*jnrC;
451 j_coord_offsetD = DIM*jnrD;
452 j_coord_offsetE = DIM*jnrE;
453 j_coord_offsetF = DIM*jnrF;
454 j_coord_offsetG = DIM*jnrG;
455 j_coord_offsetH = DIM*jnrH;
457 /* load j atom coordinates */
458 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
459 x+j_coord_offsetC,x+j_coord_offsetD,
460 x+j_coord_offsetE,x+j_coord_offsetF,
461 x+j_coord_offsetG,x+j_coord_offsetH,
464 /* Calculate displacement vector */
465 dx10 = _mm256_sub_ps(ix1,jx0);
466 dy10 = _mm256_sub_ps(iy1,jy0);
467 dz10 = _mm256_sub_ps(iz1,jz0);
468 dx20 = _mm256_sub_ps(ix2,jx0);
469 dy20 = _mm256_sub_ps(iy2,jy0);
470 dz20 = _mm256_sub_ps(iz2,jz0);
471 dx30 = _mm256_sub_ps(ix3,jx0);
472 dy30 = _mm256_sub_ps(iy3,jy0);
473 dz30 = _mm256_sub_ps(iz3,jz0);
475 /* Calculate squared distance and things based on it */
476 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
477 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
478 rsq30 = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
480 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
481 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
482 rinv30 = gmx_mm256_invsqrt_ps(rsq30);
484 /* Load parameters for j particles */
485 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
486 charge+jnrC+0,charge+jnrD+0,
487 charge+jnrE+0,charge+jnrF+0,
488 charge+jnrG+0,charge+jnrH+0);
490 fjx0 = _mm256_setzero_ps();
491 fjy0 = _mm256_setzero_ps();
492 fjz0 = _mm256_setzero_ps();
494 /**************************
495 * CALCULATE INTERACTIONS *
496 **************************/
498 r10 = _mm256_mul_ps(rsq10,rinv10);
499 r10 = _mm256_andnot_ps(dummy_mask,r10);
501 /* Compute parameters for interactions between i and j atoms */
502 qq10 = _mm256_mul_ps(iq1,jq0);
504 /* Calculate table index by multiplying r with table scale and truncate to integer */
505 rt = _mm256_mul_ps(r10,vftabscale);
506 vfitab = _mm256_cvttps_epi32(rt);
507 vfeps = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
508 /* AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
509 vfitab_lo = _mm256_extractf128_si256(vfitab,0x0);
510 vfitab_hi = _mm256_extractf128_si256(vfitab,0x1);
511 vfitab_lo = _mm_slli_epi32(vfitab_lo,2);
512 vfitab_hi = _mm_slli_epi32(vfitab_hi,2);
514 /* CUBIC SPLINE TABLE ELECTROSTATICS */
515 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
516 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
517 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
518 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
519 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
520 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
521 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
522 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
523 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
524 Heps = _mm256_mul_ps(vfeps,H);
525 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
526 VV = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
527 velec = _mm256_mul_ps(qq10,VV);
528 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
529 felec = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
531 /* Update potential sum for this i atom from the interaction with this j atom. */
532 velec = _mm256_andnot_ps(dummy_mask,velec);
533 velecsum = _mm256_add_ps(velecsum,velec);
537 fscal = _mm256_andnot_ps(dummy_mask,fscal);
539 /* Calculate temporary vectorial force */
540 tx = _mm256_mul_ps(fscal,dx10);
541 ty = _mm256_mul_ps(fscal,dy10);
542 tz = _mm256_mul_ps(fscal,dz10);
544 /* Update vectorial force */
545 fix1 = _mm256_add_ps(fix1,tx);
546 fiy1 = _mm256_add_ps(fiy1,ty);
547 fiz1 = _mm256_add_ps(fiz1,tz);
549 fjx0 = _mm256_add_ps(fjx0,tx);
550 fjy0 = _mm256_add_ps(fjy0,ty);
551 fjz0 = _mm256_add_ps(fjz0,tz);
553 /**************************
554 * CALCULATE INTERACTIONS *
555 **************************/
557 r20 = _mm256_mul_ps(rsq20,rinv20);
558 r20 = _mm256_andnot_ps(dummy_mask,r20);
560 /* Compute parameters for interactions between i and j atoms */
561 qq20 = _mm256_mul_ps(iq2,jq0);
563 /* Calculate table index by multiplying r with table scale and truncate to integer */
564 rt = _mm256_mul_ps(r20,vftabscale);
565 vfitab = _mm256_cvttps_epi32(rt);
566 vfeps = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
567 /* AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
568 vfitab_lo = _mm256_extractf128_si256(vfitab,0x0);
569 vfitab_hi = _mm256_extractf128_si256(vfitab,0x1);
570 vfitab_lo = _mm_slli_epi32(vfitab_lo,2);
571 vfitab_hi = _mm_slli_epi32(vfitab_hi,2);
573 /* CUBIC SPLINE TABLE ELECTROSTATICS */
574 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
575 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
576 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
577 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
578 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
579 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
580 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
581 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
582 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
583 Heps = _mm256_mul_ps(vfeps,H);
584 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
585 VV = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
586 velec = _mm256_mul_ps(qq20,VV);
587 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
588 felec = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
590 /* Update potential sum for this i atom from the interaction with this j atom. */
591 velec = _mm256_andnot_ps(dummy_mask,velec);
592 velecsum = _mm256_add_ps(velecsum,velec);
596 fscal = _mm256_andnot_ps(dummy_mask,fscal);
598 /* Calculate temporary vectorial force */
599 tx = _mm256_mul_ps(fscal,dx20);
600 ty = _mm256_mul_ps(fscal,dy20);
601 tz = _mm256_mul_ps(fscal,dz20);
603 /* Update vectorial force */
604 fix2 = _mm256_add_ps(fix2,tx);
605 fiy2 = _mm256_add_ps(fiy2,ty);
606 fiz2 = _mm256_add_ps(fiz2,tz);
608 fjx0 = _mm256_add_ps(fjx0,tx);
609 fjy0 = _mm256_add_ps(fjy0,ty);
610 fjz0 = _mm256_add_ps(fjz0,tz);
612 /**************************
613 * CALCULATE INTERACTIONS *
614 **************************/
616 r30 = _mm256_mul_ps(rsq30,rinv30);
617 r30 = _mm256_andnot_ps(dummy_mask,r30);
619 /* Compute parameters for interactions between i and j atoms */
620 qq30 = _mm256_mul_ps(iq3,jq0);
622 /* Calculate table index by multiplying r with table scale and truncate to integer */
623 rt = _mm256_mul_ps(r30,vftabscale);
624 vfitab = _mm256_cvttps_epi32(rt);
625 vfeps = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
626 /* AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
627 vfitab_lo = _mm256_extractf128_si256(vfitab,0x0);
628 vfitab_hi = _mm256_extractf128_si256(vfitab,0x1);
629 vfitab_lo = _mm_slli_epi32(vfitab_lo,2);
630 vfitab_hi = _mm_slli_epi32(vfitab_hi,2);
632 /* CUBIC SPLINE TABLE ELECTROSTATICS */
633 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
634 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
635 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
636 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
637 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
638 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
639 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
640 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
641 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
642 Heps = _mm256_mul_ps(vfeps,H);
643 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
644 VV = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
645 velec = _mm256_mul_ps(qq30,VV);
646 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
647 felec = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq30,FF),_mm256_mul_ps(vftabscale,rinv30)));
649 /* Update potential sum for this i atom from the interaction with this j atom. */
650 velec = _mm256_andnot_ps(dummy_mask,velec);
651 velecsum = _mm256_add_ps(velecsum,velec);
655 fscal = _mm256_andnot_ps(dummy_mask,fscal);
657 /* Calculate temporary vectorial force */
658 tx = _mm256_mul_ps(fscal,dx30);
659 ty = _mm256_mul_ps(fscal,dy30);
660 tz = _mm256_mul_ps(fscal,dz30);
662 /* Update vectorial force */
663 fix3 = _mm256_add_ps(fix3,tx);
664 fiy3 = _mm256_add_ps(fiy3,ty);
665 fiz3 = _mm256_add_ps(fiz3,tz);
667 fjx0 = _mm256_add_ps(fjx0,tx);
668 fjy0 = _mm256_add_ps(fjy0,ty);
669 fjz0 = _mm256_add_ps(fjz0,tz);
671 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
672 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
673 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
674 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
675 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
676 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
677 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
678 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
680 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
682 /* Inner loop uses 135 flops */
685 /* End of innermost loop */
687 gmx_mm256_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
688 f+i_coord_offset+DIM,fshift+i_shift_offset);
691 /* Update potential energies */
692 gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
694 /* Increment number of inner iterations */
695 inneriter += j_index_end - j_index_start;
697 /* Outer loop uses 19 flops */
700 /* Increment number of outer iterations */
703 /* Update outer/inner flops */
705 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*135);
708 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwNone_GeomW4P1_F_avx_256_single
709 * Electrostatics interaction: CubicSplineTable
710 * VdW interaction: None
711 * Geometry: Water4-Particle
712 * Calculate force/pot: Force
715 nb_kernel_ElecCSTab_VdwNone_GeomW4P1_F_avx_256_single
716 (t_nblist * gmx_restrict nlist,
717 rvec * gmx_restrict xx,
718 rvec * gmx_restrict ff,
719 t_forcerec * gmx_restrict fr,
720 t_mdatoms * gmx_restrict mdatoms,
721 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
722 t_nrnb * gmx_restrict nrnb)
724 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
725 * just 0 for non-waters.
726 * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
727 * jnr indices corresponding to data put in the four positions in the SIMD register.
729 int i_shift_offset,i_coord_offset,outeriter,inneriter;
730 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
731 int jnrA,jnrB,jnrC,jnrD;
732 int jnrE,jnrF,jnrG,jnrH;
733 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
734 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
735 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
736 int j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
737 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
739 real *shiftvec,*fshift,*x,*f;
740 real *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
742 __m256 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
743 real * vdwioffsetptr1;
744 __m256 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
745 real * vdwioffsetptr2;
746 __m256 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
747 real * vdwioffsetptr3;
748 __m256 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
749 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
750 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
751 __m256 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
752 __m256 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
753 __m256 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
754 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
757 __m128i vfitab_lo,vfitab_hi;
758 __m128i ifour = _mm_set1_epi32(4);
759 __m256 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
761 __m256 dummy_mask,cutoff_mask;
762 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
763 __m256 one = _mm256_set1_ps(1.0);
764 __m256 two = _mm256_set1_ps(2.0);
770 jindex = nlist->jindex;
772 shiftidx = nlist->shift;
774 shiftvec = fr->shift_vec[0];
775 fshift = fr->fshift[0];
776 facel = _mm256_set1_ps(fr->epsfac);
777 charge = mdatoms->chargeA;
779 vftab = kernel_data->table_elec->data;
780 vftabscale = _mm256_set1_ps(kernel_data->table_elec->scale);
782 /* Setup water-specific parameters */
783 inr = nlist->iinr[0];
784 iq1 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
785 iq2 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
786 iq3 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
788 /* Avoid stupid compiler warnings */
789 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
802 for(iidx=0;iidx<4*DIM;iidx++)
807 /* Start outer loop over neighborlists */
808 for(iidx=0; iidx<nri; iidx++)
810 /* Load shift vector for this list */
811 i_shift_offset = DIM*shiftidx[iidx];
813 /* Load limits for loop over neighbors */
814 j_index_start = jindex[iidx];
815 j_index_end = jindex[iidx+1];
817 /* Get outer coordinate index */
819 i_coord_offset = DIM*inr;
821 /* Load i particle coords and add shift vector */
822 gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
823 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
825 fix1 = _mm256_setzero_ps();
826 fiy1 = _mm256_setzero_ps();
827 fiz1 = _mm256_setzero_ps();
828 fix2 = _mm256_setzero_ps();
829 fiy2 = _mm256_setzero_ps();
830 fiz2 = _mm256_setzero_ps();
831 fix3 = _mm256_setzero_ps();
832 fiy3 = _mm256_setzero_ps();
833 fiz3 = _mm256_setzero_ps();
835 /* Start inner kernel loop */
836 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
839 /* Get j neighbor index, and coordinate index */
848 j_coord_offsetA = DIM*jnrA;
849 j_coord_offsetB = DIM*jnrB;
850 j_coord_offsetC = DIM*jnrC;
851 j_coord_offsetD = DIM*jnrD;
852 j_coord_offsetE = DIM*jnrE;
853 j_coord_offsetF = DIM*jnrF;
854 j_coord_offsetG = DIM*jnrG;
855 j_coord_offsetH = DIM*jnrH;
857 /* load j atom coordinates */
858 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
859 x+j_coord_offsetC,x+j_coord_offsetD,
860 x+j_coord_offsetE,x+j_coord_offsetF,
861 x+j_coord_offsetG,x+j_coord_offsetH,
864 /* Calculate displacement vector */
865 dx10 = _mm256_sub_ps(ix1,jx0);
866 dy10 = _mm256_sub_ps(iy1,jy0);
867 dz10 = _mm256_sub_ps(iz1,jz0);
868 dx20 = _mm256_sub_ps(ix2,jx0);
869 dy20 = _mm256_sub_ps(iy2,jy0);
870 dz20 = _mm256_sub_ps(iz2,jz0);
871 dx30 = _mm256_sub_ps(ix3,jx0);
872 dy30 = _mm256_sub_ps(iy3,jy0);
873 dz30 = _mm256_sub_ps(iz3,jz0);
875 /* Calculate squared distance and things based on it */
876 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
877 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
878 rsq30 = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
880 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
881 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
882 rinv30 = gmx_mm256_invsqrt_ps(rsq30);
884 /* Load parameters for j particles */
885 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
886 charge+jnrC+0,charge+jnrD+0,
887 charge+jnrE+0,charge+jnrF+0,
888 charge+jnrG+0,charge+jnrH+0);
890 fjx0 = _mm256_setzero_ps();
891 fjy0 = _mm256_setzero_ps();
892 fjz0 = _mm256_setzero_ps();
894 /**************************
895 * CALCULATE INTERACTIONS *
896 **************************/
898 r10 = _mm256_mul_ps(rsq10,rinv10);
900 /* Compute parameters for interactions between i and j atoms */
901 qq10 = _mm256_mul_ps(iq1,jq0);
903 /* Calculate table index by multiplying r with table scale and truncate to integer */
904 rt = _mm256_mul_ps(r10,vftabscale);
905 vfitab = _mm256_cvttps_epi32(rt);
906 vfeps = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
907 /* AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
908 vfitab_lo = _mm256_extractf128_si256(vfitab,0x0);
909 vfitab_hi = _mm256_extractf128_si256(vfitab,0x1);
910 vfitab_lo = _mm_slli_epi32(vfitab_lo,2);
911 vfitab_hi = _mm_slli_epi32(vfitab_hi,2);
913 /* CUBIC SPLINE TABLE ELECTROSTATICS */
914 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
915 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
916 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
917 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
918 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
919 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
920 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
921 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
922 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
923 Heps = _mm256_mul_ps(vfeps,H);
924 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
925 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
926 felec = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
930 /* Calculate temporary vectorial force */
931 tx = _mm256_mul_ps(fscal,dx10);
932 ty = _mm256_mul_ps(fscal,dy10);
933 tz = _mm256_mul_ps(fscal,dz10);
935 /* Update vectorial force */
936 fix1 = _mm256_add_ps(fix1,tx);
937 fiy1 = _mm256_add_ps(fiy1,ty);
938 fiz1 = _mm256_add_ps(fiz1,tz);
940 fjx0 = _mm256_add_ps(fjx0,tx);
941 fjy0 = _mm256_add_ps(fjy0,ty);
942 fjz0 = _mm256_add_ps(fjz0,tz);
944 /**************************
945 * CALCULATE INTERACTIONS *
946 **************************/
948 r20 = _mm256_mul_ps(rsq20,rinv20);
950 /* Compute parameters for interactions between i and j atoms */
951 qq20 = _mm256_mul_ps(iq2,jq0);
953 /* Calculate table index by multiplying r with table scale and truncate to integer */
954 rt = _mm256_mul_ps(r20,vftabscale);
955 vfitab = _mm256_cvttps_epi32(rt);
956 vfeps = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
957 /* AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
958 vfitab_lo = _mm256_extractf128_si256(vfitab,0x0);
959 vfitab_hi = _mm256_extractf128_si256(vfitab,0x1);
960 vfitab_lo = _mm_slli_epi32(vfitab_lo,2);
961 vfitab_hi = _mm_slli_epi32(vfitab_hi,2);
963 /* CUBIC SPLINE TABLE ELECTROSTATICS */
964 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
965 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
966 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
967 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
968 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
969 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
970 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
971 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
972 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
973 Heps = _mm256_mul_ps(vfeps,H);
974 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
975 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
976 felec = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
980 /* Calculate temporary vectorial force */
981 tx = _mm256_mul_ps(fscal,dx20);
982 ty = _mm256_mul_ps(fscal,dy20);
983 tz = _mm256_mul_ps(fscal,dz20);
985 /* Update vectorial force */
986 fix2 = _mm256_add_ps(fix2,tx);
987 fiy2 = _mm256_add_ps(fiy2,ty);
988 fiz2 = _mm256_add_ps(fiz2,tz);
990 fjx0 = _mm256_add_ps(fjx0,tx);
991 fjy0 = _mm256_add_ps(fjy0,ty);
992 fjz0 = _mm256_add_ps(fjz0,tz);
994 /**************************
995 * CALCULATE INTERACTIONS *
996 **************************/
998 r30 = _mm256_mul_ps(rsq30,rinv30);
1000 /* Compute parameters for interactions between i and j atoms */
1001 qq30 = _mm256_mul_ps(iq3,jq0);
1003 /* Calculate table index by multiplying r with table scale and truncate to integer */
1004 rt = _mm256_mul_ps(r30,vftabscale);
1005 vfitab = _mm256_cvttps_epi32(rt);
1006 vfeps = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1007 /* AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1008 vfitab_lo = _mm256_extractf128_si256(vfitab,0x0);
1009 vfitab_hi = _mm256_extractf128_si256(vfitab,0x1);
1010 vfitab_lo = _mm_slli_epi32(vfitab_lo,2);
1011 vfitab_hi = _mm_slli_epi32(vfitab_hi,2);
1013 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1014 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1015 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1016 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1017 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1018 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1019 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1020 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1021 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1022 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1023 Heps = _mm256_mul_ps(vfeps,H);
1024 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1025 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1026 felec = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq30,FF),_mm256_mul_ps(vftabscale,rinv30)));
1030 /* Calculate temporary vectorial force */
1031 tx = _mm256_mul_ps(fscal,dx30);
1032 ty = _mm256_mul_ps(fscal,dy30);
1033 tz = _mm256_mul_ps(fscal,dz30);
1035 /* Update vectorial force */
1036 fix3 = _mm256_add_ps(fix3,tx);
1037 fiy3 = _mm256_add_ps(fiy3,ty);
1038 fiz3 = _mm256_add_ps(fiz3,tz);
1040 fjx0 = _mm256_add_ps(fjx0,tx);
1041 fjy0 = _mm256_add_ps(fjy0,ty);
1042 fjz0 = _mm256_add_ps(fjz0,tz);
1044 fjptrA = f+j_coord_offsetA;
1045 fjptrB = f+j_coord_offsetB;
1046 fjptrC = f+j_coord_offsetC;
1047 fjptrD = f+j_coord_offsetD;
1048 fjptrE = f+j_coord_offsetE;
1049 fjptrF = f+j_coord_offsetF;
1050 fjptrG = f+j_coord_offsetG;
1051 fjptrH = f+j_coord_offsetH;
1053 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1055 /* Inner loop uses 120 flops */
1058 if(jidx<j_index_end)
1061 /* Get j neighbor index, and coordinate index */
1062 jnrlistA = jjnr[jidx];
1063 jnrlistB = jjnr[jidx+1];
1064 jnrlistC = jjnr[jidx+2];
1065 jnrlistD = jjnr[jidx+3];
1066 jnrlistE = jjnr[jidx+4];
1067 jnrlistF = jjnr[jidx+5];
1068 jnrlistG = jjnr[jidx+6];
1069 jnrlistH = jjnr[jidx+7];
1070 /* Sign of each element will be negative for non-real atoms.
1071 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1072 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1074 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1075 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1077 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
1078 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
1079 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
1080 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
1081 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
1082 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
1083 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
1084 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
1085 j_coord_offsetA = DIM*jnrA;
1086 j_coord_offsetB = DIM*jnrB;
1087 j_coord_offsetC = DIM*jnrC;
1088 j_coord_offsetD = DIM*jnrD;
1089 j_coord_offsetE = DIM*jnrE;
1090 j_coord_offsetF = DIM*jnrF;
1091 j_coord_offsetG = DIM*jnrG;
1092 j_coord_offsetH = DIM*jnrH;
1094 /* load j atom coordinates */
1095 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1096 x+j_coord_offsetC,x+j_coord_offsetD,
1097 x+j_coord_offsetE,x+j_coord_offsetF,
1098 x+j_coord_offsetG,x+j_coord_offsetH,
1101 /* Calculate displacement vector */
1102 dx10 = _mm256_sub_ps(ix1,jx0);
1103 dy10 = _mm256_sub_ps(iy1,jy0);
1104 dz10 = _mm256_sub_ps(iz1,jz0);
1105 dx20 = _mm256_sub_ps(ix2,jx0);
1106 dy20 = _mm256_sub_ps(iy2,jy0);
1107 dz20 = _mm256_sub_ps(iz2,jz0);
1108 dx30 = _mm256_sub_ps(ix3,jx0);
1109 dy30 = _mm256_sub_ps(iy3,jy0);
1110 dz30 = _mm256_sub_ps(iz3,jz0);
1112 /* Calculate squared distance and things based on it */
1113 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1114 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1115 rsq30 = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1117 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
1118 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
1119 rinv30 = gmx_mm256_invsqrt_ps(rsq30);
1121 /* Load parameters for j particles */
1122 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1123 charge+jnrC+0,charge+jnrD+0,
1124 charge+jnrE+0,charge+jnrF+0,
1125 charge+jnrG+0,charge+jnrH+0);
1127 fjx0 = _mm256_setzero_ps();
1128 fjy0 = _mm256_setzero_ps();
1129 fjz0 = _mm256_setzero_ps();
1131 /**************************
1132 * CALCULATE INTERACTIONS *
1133 **************************/
1135 r10 = _mm256_mul_ps(rsq10,rinv10);
1136 r10 = _mm256_andnot_ps(dummy_mask,r10);
1138 /* Compute parameters for interactions between i and j atoms */
1139 qq10 = _mm256_mul_ps(iq1,jq0);
1141 /* Calculate table index by multiplying r with table scale and truncate to integer */
1142 rt = _mm256_mul_ps(r10,vftabscale);
1143 vfitab = _mm256_cvttps_epi32(rt);
1144 vfeps = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1145 /* AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1146 vfitab_lo = _mm256_extractf128_si256(vfitab,0x0);
1147 vfitab_hi = _mm256_extractf128_si256(vfitab,0x1);
1148 vfitab_lo = _mm_slli_epi32(vfitab_lo,2);
1149 vfitab_hi = _mm_slli_epi32(vfitab_hi,2);
1151 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1152 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1153 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1154 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1155 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1156 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1157 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1158 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1159 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1160 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1161 Heps = _mm256_mul_ps(vfeps,H);
1162 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1163 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1164 felec = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
1168 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1170 /* Calculate temporary vectorial force */
1171 tx = _mm256_mul_ps(fscal,dx10);
1172 ty = _mm256_mul_ps(fscal,dy10);
1173 tz = _mm256_mul_ps(fscal,dz10);
1175 /* Update vectorial force */
1176 fix1 = _mm256_add_ps(fix1,tx);
1177 fiy1 = _mm256_add_ps(fiy1,ty);
1178 fiz1 = _mm256_add_ps(fiz1,tz);
1180 fjx0 = _mm256_add_ps(fjx0,tx);
1181 fjy0 = _mm256_add_ps(fjy0,ty);
1182 fjz0 = _mm256_add_ps(fjz0,tz);
1184 /**************************
1185 * CALCULATE INTERACTIONS *
1186 **************************/
1188 r20 = _mm256_mul_ps(rsq20,rinv20);
1189 r20 = _mm256_andnot_ps(dummy_mask,r20);
1191 /* Compute parameters for interactions between i and j atoms */
1192 qq20 = _mm256_mul_ps(iq2,jq0);
1194 /* Calculate table index by multiplying r with table scale and truncate to integer */
1195 rt = _mm256_mul_ps(r20,vftabscale);
1196 vfitab = _mm256_cvttps_epi32(rt);
1197 vfeps = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1198 /* AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1199 vfitab_lo = _mm256_extractf128_si256(vfitab,0x0);
1200 vfitab_hi = _mm256_extractf128_si256(vfitab,0x1);
1201 vfitab_lo = _mm_slli_epi32(vfitab_lo,2);
1202 vfitab_hi = _mm_slli_epi32(vfitab_hi,2);
1204 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1205 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1206 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1207 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1208 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1209 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1210 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1211 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1212 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1213 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1214 Heps = _mm256_mul_ps(vfeps,H);
1215 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1216 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1217 felec = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
1221 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1223 /* Calculate temporary vectorial force */
1224 tx = _mm256_mul_ps(fscal,dx20);
1225 ty = _mm256_mul_ps(fscal,dy20);
1226 tz = _mm256_mul_ps(fscal,dz20);
1228 /* Update vectorial force */
1229 fix2 = _mm256_add_ps(fix2,tx);
1230 fiy2 = _mm256_add_ps(fiy2,ty);
1231 fiz2 = _mm256_add_ps(fiz2,tz);
1233 fjx0 = _mm256_add_ps(fjx0,tx);
1234 fjy0 = _mm256_add_ps(fjy0,ty);
1235 fjz0 = _mm256_add_ps(fjz0,tz);
1237 /**************************
1238 * CALCULATE INTERACTIONS *
1239 **************************/
1241 r30 = _mm256_mul_ps(rsq30,rinv30);
1242 r30 = _mm256_andnot_ps(dummy_mask,r30);
1244 /* Compute parameters for interactions between i and j atoms */
1245 qq30 = _mm256_mul_ps(iq3,jq0);
1247 /* Calculate table index by multiplying r with table scale and truncate to integer */
1248 rt = _mm256_mul_ps(r30,vftabscale);
1249 vfitab = _mm256_cvttps_epi32(rt);
1250 vfeps = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1251 /* AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1252 vfitab_lo = _mm256_extractf128_si256(vfitab,0x0);
1253 vfitab_hi = _mm256_extractf128_si256(vfitab,0x1);
1254 vfitab_lo = _mm_slli_epi32(vfitab_lo,2);
1255 vfitab_hi = _mm_slli_epi32(vfitab_hi,2);
1257 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1258 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1259 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1260 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1261 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1262 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1263 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1264 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1265 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1266 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1267 Heps = _mm256_mul_ps(vfeps,H);
1268 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1269 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1270 felec = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq30,FF),_mm256_mul_ps(vftabscale,rinv30)));
1274 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1276 /* Calculate temporary vectorial force */
1277 tx = _mm256_mul_ps(fscal,dx30);
1278 ty = _mm256_mul_ps(fscal,dy30);
1279 tz = _mm256_mul_ps(fscal,dz30);
1281 /* Update vectorial force */
1282 fix3 = _mm256_add_ps(fix3,tx);
1283 fiy3 = _mm256_add_ps(fiy3,ty);
1284 fiz3 = _mm256_add_ps(fiz3,tz);
1286 fjx0 = _mm256_add_ps(fjx0,tx);
1287 fjy0 = _mm256_add_ps(fjy0,ty);
1288 fjz0 = _mm256_add_ps(fjz0,tz);
1290 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1291 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1292 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1293 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1294 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1295 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1296 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1297 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1299 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1301 /* Inner loop uses 123 flops */
1304 /* End of innermost loop */
1306 gmx_mm256_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1307 f+i_coord_offset+DIM,fshift+i_shift_offset);
1309 /* Increment number of inner iterations */
1310 inneriter += j_index_end - j_index_start;
1312 /* Outer loop uses 18 flops */
1315 /* Increment number of outer iterations */
1318 /* Update outer/inner flops */
1320 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*123);