2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013, 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 "types/simple.h"
49 #include "gmx_math_x86_avx_256_single.h"
50 #include "kernelutil_x86_avx_256_single.h"
53 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwCSTab_GeomP1P1_VF_avx_256_single
54 * Electrostatics interaction: Ewald
55 * VdW interaction: CubicSplineTable
56 * Geometry: Particle-Particle
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecEw_VdwCSTab_GeomP1P1_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 * vdwioffsetptr0;
89 __m256 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
90 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
91 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
92 __m256 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
93 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
96 __m256 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
99 __m256 one_sixth = _mm256_set1_ps(1.0/6.0);
100 __m256 one_twelfth = _mm256_set1_ps(1.0/12.0);
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;
107 __m128i ewitab_lo,ewitab_hi;
108 __m256 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
109 __m256 beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
111 __m256 dummy_mask,cutoff_mask;
112 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
113 __m256 one = _mm256_set1_ps(1.0);
114 __m256 two = _mm256_set1_ps(2.0);
120 jindex = nlist->jindex;
122 shiftidx = nlist->shift;
124 shiftvec = fr->shift_vec[0];
125 fshift = fr->fshift[0];
126 facel = _mm256_set1_ps(fr->epsfac);
127 charge = mdatoms->chargeA;
128 nvdwtype = fr->ntype;
130 vdwtype = mdatoms->typeA;
132 vftab = kernel_data->table_vdw->data;
133 vftabscale = _mm256_set1_ps(kernel_data->table_vdw->scale);
135 sh_ewald = _mm256_set1_ps(fr->ic->sh_ewald);
136 beta = _mm256_set1_ps(fr->ic->ewaldcoeff);
137 beta2 = _mm256_mul_ps(beta,beta);
138 beta3 = _mm256_mul_ps(beta,beta2);
140 ewtab = fr->ic->tabq_coul_FDV0;
141 ewtabscale = _mm256_set1_ps(fr->ic->tabq_scale);
142 ewtabhalfspace = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
144 /* Avoid stupid compiler warnings */
145 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
158 for(iidx=0;iidx<4*DIM;iidx++)
163 /* Start outer loop over neighborlists */
164 for(iidx=0; iidx<nri; iidx++)
166 /* Load shift vector for this list */
167 i_shift_offset = DIM*shiftidx[iidx];
169 /* Load limits for loop over neighbors */
170 j_index_start = jindex[iidx];
171 j_index_end = jindex[iidx+1];
173 /* Get outer coordinate index */
175 i_coord_offset = DIM*inr;
177 /* Load i particle coords and add shift vector */
178 gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
180 fix0 = _mm256_setzero_ps();
181 fiy0 = _mm256_setzero_ps();
182 fiz0 = _mm256_setzero_ps();
184 /* Load parameters for i particles */
185 iq0 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
186 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
188 /* Reset potential sums */
189 velecsum = _mm256_setzero_ps();
190 vvdwsum = _mm256_setzero_ps();
192 /* Start inner kernel loop */
193 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
196 /* Get j neighbor index, and coordinate index */
205 j_coord_offsetA = DIM*jnrA;
206 j_coord_offsetB = DIM*jnrB;
207 j_coord_offsetC = DIM*jnrC;
208 j_coord_offsetD = DIM*jnrD;
209 j_coord_offsetE = DIM*jnrE;
210 j_coord_offsetF = DIM*jnrF;
211 j_coord_offsetG = DIM*jnrG;
212 j_coord_offsetH = DIM*jnrH;
214 /* load j atom coordinates */
215 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
216 x+j_coord_offsetC,x+j_coord_offsetD,
217 x+j_coord_offsetE,x+j_coord_offsetF,
218 x+j_coord_offsetG,x+j_coord_offsetH,
221 /* Calculate displacement vector */
222 dx00 = _mm256_sub_ps(ix0,jx0);
223 dy00 = _mm256_sub_ps(iy0,jy0);
224 dz00 = _mm256_sub_ps(iz0,jz0);
226 /* Calculate squared distance and things based on it */
227 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
229 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
231 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
233 /* Load parameters for j particles */
234 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
235 charge+jnrC+0,charge+jnrD+0,
236 charge+jnrE+0,charge+jnrF+0,
237 charge+jnrG+0,charge+jnrH+0);
238 vdwjidx0A = 2*vdwtype[jnrA+0];
239 vdwjidx0B = 2*vdwtype[jnrB+0];
240 vdwjidx0C = 2*vdwtype[jnrC+0];
241 vdwjidx0D = 2*vdwtype[jnrD+0];
242 vdwjidx0E = 2*vdwtype[jnrE+0];
243 vdwjidx0F = 2*vdwtype[jnrF+0];
244 vdwjidx0G = 2*vdwtype[jnrG+0];
245 vdwjidx0H = 2*vdwtype[jnrH+0];
247 /**************************
248 * CALCULATE INTERACTIONS *
249 **************************/
251 r00 = _mm256_mul_ps(rsq00,rinv00);
253 /* Compute parameters for interactions between i and j atoms */
254 qq00 = _mm256_mul_ps(iq0,jq0);
255 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
256 vdwioffsetptr0+vdwjidx0B,
257 vdwioffsetptr0+vdwjidx0C,
258 vdwioffsetptr0+vdwjidx0D,
259 vdwioffsetptr0+vdwjidx0E,
260 vdwioffsetptr0+vdwjidx0F,
261 vdwioffsetptr0+vdwjidx0G,
262 vdwioffsetptr0+vdwjidx0H,
265 /* Calculate table index by multiplying r with table scale and truncate to integer */
266 rt = _mm256_mul_ps(r00,vftabscale);
267 vfitab = _mm256_cvttps_epi32(rt);
268 vfeps = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
269 /* AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
270 vfitab_lo = _mm256_extractf128_si256(vfitab,0x0);
271 vfitab_hi = _mm256_extractf128_si256(vfitab,0x1);
272 vfitab_lo = _mm_slli_epi32(vfitab_lo,3);
273 vfitab_hi = _mm_slli_epi32(vfitab_hi,3);
275 /* EWALD ELECTROSTATICS */
277 /* Analytical PME correction */
278 zeta2 = _mm256_mul_ps(beta2,rsq00);
279 rinv3 = _mm256_mul_ps(rinvsq00,rinv00);
280 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
281 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
282 felec = _mm256_mul_ps(qq00,felec);
283 pmecorrV = gmx_mm256_pmecorrV_ps(zeta2);
284 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
285 velec = _mm256_sub_ps(rinv00,pmecorrV);
286 velec = _mm256_mul_ps(qq00,velec);
288 /* CUBIC SPLINE TABLE DISPERSION */
289 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
290 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
291 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
292 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
293 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
294 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
295 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
296 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
297 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
298 Heps = _mm256_mul_ps(vfeps,H);
299 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
300 VV = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
301 vvdw6 = _mm256_mul_ps(c6_00,VV);
302 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
303 fvdw6 = _mm256_mul_ps(c6_00,FF);
305 /* CUBIC SPLINE TABLE REPULSION */
306 vfitab_lo = _mm_add_epi32(vfitab_lo,ifour);
307 vfitab_hi = _mm_add_epi32(vfitab_hi,ifour);
308 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
309 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
310 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
311 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
312 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
313 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
314 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
315 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
316 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
317 Heps = _mm256_mul_ps(vfeps,H);
318 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
319 VV = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
320 vvdw12 = _mm256_mul_ps(c12_00,VV);
321 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
322 fvdw12 = _mm256_mul_ps(c12_00,FF);
323 vvdw = _mm256_add_ps(vvdw12,vvdw6);
324 fvdw = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
326 /* Update potential sum for this i atom from the interaction with this j atom. */
327 velecsum = _mm256_add_ps(velecsum,velec);
328 vvdwsum = _mm256_add_ps(vvdwsum,vvdw);
330 fscal = _mm256_add_ps(felec,fvdw);
332 /* Calculate temporary vectorial force */
333 tx = _mm256_mul_ps(fscal,dx00);
334 ty = _mm256_mul_ps(fscal,dy00);
335 tz = _mm256_mul_ps(fscal,dz00);
337 /* Update vectorial force */
338 fix0 = _mm256_add_ps(fix0,tx);
339 fiy0 = _mm256_add_ps(fiy0,ty);
340 fiz0 = _mm256_add_ps(fiz0,tz);
342 fjptrA = f+j_coord_offsetA;
343 fjptrB = f+j_coord_offsetB;
344 fjptrC = f+j_coord_offsetC;
345 fjptrD = f+j_coord_offsetD;
346 fjptrE = f+j_coord_offsetE;
347 fjptrF = f+j_coord_offsetF;
348 fjptrG = f+j_coord_offsetG;
349 fjptrH = f+j_coord_offsetH;
350 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
352 /* Inner loop uses 118 flops */
358 /* Get j neighbor index, and coordinate index */
359 jnrlistA = jjnr[jidx];
360 jnrlistB = jjnr[jidx+1];
361 jnrlistC = jjnr[jidx+2];
362 jnrlistD = jjnr[jidx+3];
363 jnrlistE = jjnr[jidx+4];
364 jnrlistF = jjnr[jidx+5];
365 jnrlistG = jjnr[jidx+6];
366 jnrlistH = jjnr[jidx+7];
367 /* Sign of each element will be negative for non-real atoms.
368 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
369 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
371 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
372 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
374 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
375 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
376 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
377 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
378 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
379 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
380 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
381 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
382 j_coord_offsetA = DIM*jnrA;
383 j_coord_offsetB = DIM*jnrB;
384 j_coord_offsetC = DIM*jnrC;
385 j_coord_offsetD = DIM*jnrD;
386 j_coord_offsetE = DIM*jnrE;
387 j_coord_offsetF = DIM*jnrF;
388 j_coord_offsetG = DIM*jnrG;
389 j_coord_offsetH = DIM*jnrH;
391 /* load j atom coordinates */
392 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
393 x+j_coord_offsetC,x+j_coord_offsetD,
394 x+j_coord_offsetE,x+j_coord_offsetF,
395 x+j_coord_offsetG,x+j_coord_offsetH,
398 /* Calculate displacement vector */
399 dx00 = _mm256_sub_ps(ix0,jx0);
400 dy00 = _mm256_sub_ps(iy0,jy0);
401 dz00 = _mm256_sub_ps(iz0,jz0);
403 /* Calculate squared distance and things based on it */
404 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
406 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
408 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
410 /* Load parameters for j particles */
411 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
412 charge+jnrC+0,charge+jnrD+0,
413 charge+jnrE+0,charge+jnrF+0,
414 charge+jnrG+0,charge+jnrH+0);
415 vdwjidx0A = 2*vdwtype[jnrA+0];
416 vdwjidx0B = 2*vdwtype[jnrB+0];
417 vdwjidx0C = 2*vdwtype[jnrC+0];
418 vdwjidx0D = 2*vdwtype[jnrD+0];
419 vdwjidx0E = 2*vdwtype[jnrE+0];
420 vdwjidx0F = 2*vdwtype[jnrF+0];
421 vdwjidx0G = 2*vdwtype[jnrG+0];
422 vdwjidx0H = 2*vdwtype[jnrH+0];
424 /**************************
425 * CALCULATE INTERACTIONS *
426 **************************/
428 r00 = _mm256_mul_ps(rsq00,rinv00);
429 r00 = _mm256_andnot_ps(dummy_mask,r00);
431 /* Compute parameters for interactions between i and j atoms */
432 qq00 = _mm256_mul_ps(iq0,jq0);
433 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
434 vdwioffsetptr0+vdwjidx0B,
435 vdwioffsetptr0+vdwjidx0C,
436 vdwioffsetptr0+vdwjidx0D,
437 vdwioffsetptr0+vdwjidx0E,
438 vdwioffsetptr0+vdwjidx0F,
439 vdwioffsetptr0+vdwjidx0G,
440 vdwioffsetptr0+vdwjidx0H,
443 /* Calculate table index by multiplying r with table scale and truncate to integer */
444 rt = _mm256_mul_ps(r00,vftabscale);
445 vfitab = _mm256_cvttps_epi32(rt);
446 vfeps = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
447 /* AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
448 vfitab_lo = _mm256_extractf128_si256(vfitab,0x0);
449 vfitab_hi = _mm256_extractf128_si256(vfitab,0x1);
450 vfitab_lo = _mm_slli_epi32(vfitab_lo,3);
451 vfitab_hi = _mm_slli_epi32(vfitab_hi,3);
453 /* EWALD ELECTROSTATICS */
455 /* Analytical PME correction */
456 zeta2 = _mm256_mul_ps(beta2,rsq00);
457 rinv3 = _mm256_mul_ps(rinvsq00,rinv00);
458 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
459 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
460 felec = _mm256_mul_ps(qq00,felec);
461 pmecorrV = gmx_mm256_pmecorrV_ps(zeta2);
462 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
463 velec = _mm256_sub_ps(rinv00,pmecorrV);
464 velec = _mm256_mul_ps(qq00,velec);
466 /* CUBIC SPLINE TABLE DISPERSION */
467 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
468 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
469 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
470 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
471 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
472 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
473 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
474 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
475 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
476 Heps = _mm256_mul_ps(vfeps,H);
477 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
478 VV = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
479 vvdw6 = _mm256_mul_ps(c6_00,VV);
480 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
481 fvdw6 = _mm256_mul_ps(c6_00,FF);
483 /* CUBIC SPLINE TABLE REPULSION */
484 vfitab_lo = _mm_add_epi32(vfitab_lo,ifour);
485 vfitab_hi = _mm_add_epi32(vfitab_hi,ifour);
486 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
487 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
488 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
489 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
490 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
491 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
492 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
493 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
494 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
495 Heps = _mm256_mul_ps(vfeps,H);
496 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
497 VV = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
498 vvdw12 = _mm256_mul_ps(c12_00,VV);
499 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
500 fvdw12 = _mm256_mul_ps(c12_00,FF);
501 vvdw = _mm256_add_ps(vvdw12,vvdw6);
502 fvdw = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
504 /* Update potential sum for this i atom from the interaction with this j atom. */
505 velec = _mm256_andnot_ps(dummy_mask,velec);
506 velecsum = _mm256_add_ps(velecsum,velec);
507 vvdw = _mm256_andnot_ps(dummy_mask,vvdw);
508 vvdwsum = _mm256_add_ps(vvdwsum,vvdw);
510 fscal = _mm256_add_ps(felec,fvdw);
512 fscal = _mm256_andnot_ps(dummy_mask,fscal);
514 /* Calculate temporary vectorial force */
515 tx = _mm256_mul_ps(fscal,dx00);
516 ty = _mm256_mul_ps(fscal,dy00);
517 tz = _mm256_mul_ps(fscal,dz00);
519 /* Update vectorial force */
520 fix0 = _mm256_add_ps(fix0,tx);
521 fiy0 = _mm256_add_ps(fiy0,ty);
522 fiz0 = _mm256_add_ps(fiz0,tz);
524 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
525 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
526 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
527 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
528 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
529 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
530 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
531 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
532 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
534 /* Inner loop uses 119 flops */
537 /* End of innermost loop */
539 gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
540 f+i_coord_offset,fshift+i_shift_offset);
543 /* Update potential energies */
544 gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
545 gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
547 /* Increment number of inner iterations */
548 inneriter += j_index_end - j_index_start;
550 /* Outer loop uses 9 flops */
553 /* Increment number of outer iterations */
556 /* Update outer/inner flops */
558 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*119);
561 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwCSTab_GeomP1P1_F_avx_256_single
562 * Electrostatics interaction: Ewald
563 * VdW interaction: CubicSplineTable
564 * Geometry: Particle-Particle
565 * Calculate force/pot: Force
568 nb_kernel_ElecEw_VdwCSTab_GeomP1P1_F_avx_256_single
569 (t_nblist * gmx_restrict nlist,
570 rvec * gmx_restrict xx,
571 rvec * gmx_restrict ff,
572 t_forcerec * gmx_restrict fr,
573 t_mdatoms * gmx_restrict mdatoms,
574 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
575 t_nrnb * gmx_restrict nrnb)
577 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
578 * just 0 for non-waters.
579 * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
580 * jnr indices corresponding to data put in the four positions in the SIMD register.
582 int i_shift_offset,i_coord_offset,outeriter,inneriter;
583 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
584 int jnrA,jnrB,jnrC,jnrD;
585 int jnrE,jnrF,jnrG,jnrH;
586 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
587 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
588 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
589 int j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
590 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
592 real *shiftvec,*fshift,*x,*f;
593 real *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
595 __m256 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
596 real * vdwioffsetptr0;
597 __m256 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
598 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
599 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
600 __m256 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
601 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
604 __m256 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
607 __m256 one_sixth = _mm256_set1_ps(1.0/6.0);
608 __m256 one_twelfth = _mm256_set1_ps(1.0/12.0);
610 __m128i vfitab_lo,vfitab_hi;
611 __m128i ifour = _mm_set1_epi32(4);
612 __m256 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
615 __m128i ewitab_lo,ewitab_hi;
616 __m256 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
617 __m256 beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
619 __m256 dummy_mask,cutoff_mask;
620 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
621 __m256 one = _mm256_set1_ps(1.0);
622 __m256 two = _mm256_set1_ps(2.0);
628 jindex = nlist->jindex;
630 shiftidx = nlist->shift;
632 shiftvec = fr->shift_vec[0];
633 fshift = fr->fshift[0];
634 facel = _mm256_set1_ps(fr->epsfac);
635 charge = mdatoms->chargeA;
636 nvdwtype = fr->ntype;
638 vdwtype = mdatoms->typeA;
640 vftab = kernel_data->table_vdw->data;
641 vftabscale = _mm256_set1_ps(kernel_data->table_vdw->scale);
643 sh_ewald = _mm256_set1_ps(fr->ic->sh_ewald);
644 beta = _mm256_set1_ps(fr->ic->ewaldcoeff);
645 beta2 = _mm256_mul_ps(beta,beta);
646 beta3 = _mm256_mul_ps(beta,beta2);
648 ewtab = fr->ic->tabq_coul_F;
649 ewtabscale = _mm256_set1_ps(fr->ic->tabq_scale);
650 ewtabhalfspace = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
652 /* Avoid stupid compiler warnings */
653 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
666 for(iidx=0;iidx<4*DIM;iidx++)
671 /* Start outer loop over neighborlists */
672 for(iidx=0; iidx<nri; iidx++)
674 /* Load shift vector for this list */
675 i_shift_offset = DIM*shiftidx[iidx];
677 /* Load limits for loop over neighbors */
678 j_index_start = jindex[iidx];
679 j_index_end = jindex[iidx+1];
681 /* Get outer coordinate index */
683 i_coord_offset = DIM*inr;
685 /* Load i particle coords and add shift vector */
686 gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
688 fix0 = _mm256_setzero_ps();
689 fiy0 = _mm256_setzero_ps();
690 fiz0 = _mm256_setzero_ps();
692 /* Load parameters for i particles */
693 iq0 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
694 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
696 /* Start inner kernel loop */
697 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
700 /* Get j neighbor index, and coordinate index */
709 j_coord_offsetA = DIM*jnrA;
710 j_coord_offsetB = DIM*jnrB;
711 j_coord_offsetC = DIM*jnrC;
712 j_coord_offsetD = DIM*jnrD;
713 j_coord_offsetE = DIM*jnrE;
714 j_coord_offsetF = DIM*jnrF;
715 j_coord_offsetG = DIM*jnrG;
716 j_coord_offsetH = DIM*jnrH;
718 /* load j atom coordinates */
719 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
720 x+j_coord_offsetC,x+j_coord_offsetD,
721 x+j_coord_offsetE,x+j_coord_offsetF,
722 x+j_coord_offsetG,x+j_coord_offsetH,
725 /* Calculate displacement vector */
726 dx00 = _mm256_sub_ps(ix0,jx0);
727 dy00 = _mm256_sub_ps(iy0,jy0);
728 dz00 = _mm256_sub_ps(iz0,jz0);
730 /* Calculate squared distance and things based on it */
731 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
733 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
735 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
737 /* Load parameters for j particles */
738 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
739 charge+jnrC+0,charge+jnrD+0,
740 charge+jnrE+0,charge+jnrF+0,
741 charge+jnrG+0,charge+jnrH+0);
742 vdwjidx0A = 2*vdwtype[jnrA+0];
743 vdwjidx0B = 2*vdwtype[jnrB+0];
744 vdwjidx0C = 2*vdwtype[jnrC+0];
745 vdwjidx0D = 2*vdwtype[jnrD+0];
746 vdwjidx0E = 2*vdwtype[jnrE+0];
747 vdwjidx0F = 2*vdwtype[jnrF+0];
748 vdwjidx0G = 2*vdwtype[jnrG+0];
749 vdwjidx0H = 2*vdwtype[jnrH+0];
751 /**************************
752 * CALCULATE INTERACTIONS *
753 **************************/
755 r00 = _mm256_mul_ps(rsq00,rinv00);
757 /* Compute parameters for interactions between i and j atoms */
758 qq00 = _mm256_mul_ps(iq0,jq0);
759 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
760 vdwioffsetptr0+vdwjidx0B,
761 vdwioffsetptr0+vdwjidx0C,
762 vdwioffsetptr0+vdwjidx0D,
763 vdwioffsetptr0+vdwjidx0E,
764 vdwioffsetptr0+vdwjidx0F,
765 vdwioffsetptr0+vdwjidx0G,
766 vdwioffsetptr0+vdwjidx0H,
769 /* Calculate table index by multiplying r with table scale and truncate to integer */
770 rt = _mm256_mul_ps(r00,vftabscale);
771 vfitab = _mm256_cvttps_epi32(rt);
772 vfeps = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
773 /* AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
774 vfitab_lo = _mm256_extractf128_si256(vfitab,0x0);
775 vfitab_hi = _mm256_extractf128_si256(vfitab,0x1);
776 vfitab_lo = _mm_slli_epi32(vfitab_lo,3);
777 vfitab_hi = _mm_slli_epi32(vfitab_hi,3);
779 /* EWALD ELECTROSTATICS */
781 /* Analytical PME correction */
782 zeta2 = _mm256_mul_ps(beta2,rsq00);
783 rinv3 = _mm256_mul_ps(rinvsq00,rinv00);
784 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
785 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
786 felec = _mm256_mul_ps(qq00,felec);
788 /* CUBIC SPLINE TABLE DISPERSION */
789 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
790 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
791 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
792 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
793 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
794 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
795 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
796 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
797 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
798 Heps = _mm256_mul_ps(vfeps,H);
799 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
800 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
801 fvdw6 = _mm256_mul_ps(c6_00,FF);
803 /* CUBIC SPLINE TABLE REPULSION */
804 vfitab_lo = _mm_add_epi32(vfitab_lo,ifour);
805 vfitab_hi = _mm_add_epi32(vfitab_hi,ifour);
806 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
807 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
808 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
809 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
810 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
811 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
812 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
813 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
814 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
815 Heps = _mm256_mul_ps(vfeps,H);
816 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
817 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
818 fvdw12 = _mm256_mul_ps(c12_00,FF);
819 fvdw = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
821 fscal = _mm256_add_ps(felec,fvdw);
823 /* Calculate temporary vectorial force */
824 tx = _mm256_mul_ps(fscal,dx00);
825 ty = _mm256_mul_ps(fscal,dy00);
826 tz = _mm256_mul_ps(fscal,dz00);
828 /* Update vectorial force */
829 fix0 = _mm256_add_ps(fix0,tx);
830 fiy0 = _mm256_add_ps(fiy0,ty);
831 fiz0 = _mm256_add_ps(fiz0,tz);
833 fjptrA = f+j_coord_offsetA;
834 fjptrB = f+j_coord_offsetB;
835 fjptrC = f+j_coord_offsetC;
836 fjptrD = f+j_coord_offsetD;
837 fjptrE = f+j_coord_offsetE;
838 fjptrF = f+j_coord_offsetF;
839 fjptrG = f+j_coord_offsetG;
840 fjptrH = f+j_coord_offsetH;
841 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
843 /* Inner loop uses 82 flops */
849 /* Get j neighbor index, and coordinate index */
850 jnrlistA = jjnr[jidx];
851 jnrlistB = jjnr[jidx+1];
852 jnrlistC = jjnr[jidx+2];
853 jnrlistD = jjnr[jidx+3];
854 jnrlistE = jjnr[jidx+4];
855 jnrlistF = jjnr[jidx+5];
856 jnrlistG = jjnr[jidx+6];
857 jnrlistH = jjnr[jidx+7];
858 /* Sign of each element will be negative for non-real atoms.
859 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
860 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
862 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
863 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
865 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
866 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
867 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
868 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
869 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
870 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
871 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
872 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
873 j_coord_offsetA = DIM*jnrA;
874 j_coord_offsetB = DIM*jnrB;
875 j_coord_offsetC = DIM*jnrC;
876 j_coord_offsetD = DIM*jnrD;
877 j_coord_offsetE = DIM*jnrE;
878 j_coord_offsetF = DIM*jnrF;
879 j_coord_offsetG = DIM*jnrG;
880 j_coord_offsetH = DIM*jnrH;
882 /* load j atom coordinates */
883 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
884 x+j_coord_offsetC,x+j_coord_offsetD,
885 x+j_coord_offsetE,x+j_coord_offsetF,
886 x+j_coord_offsetG,x+j_coord_offsetH,
889 /* Calculate displacement vector */
890 dx00 = _mm256_sub_ps(ix0,jx0);
891 dy00 = _mm256_sub_ps(iy0,jy0);
892 dz00 = _mm256_sub_ps(iz0,jz0);
894 /* Calculate squared distance and things based on it */
895 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
897 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
899 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
901 /* Load parameters for j particles */
902 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
903 charge+jnrC+0,charge+jnrD+0,
904 charge+jnrE+0,charge+jnrF+0,
905 charge+jnrG+0,charge+jnrH+0);
906 vdwjidx0A = 2*vdwtype[jnrA+0];
907 vdwjidx0B = 2*vdwtype[jnrB+0];
908 vdwjidx0C = 2*vdwtype[jnrC+0];
909 vdwjidx0D = 2*vdwtype[jnrD+0];
910 vdwjidx0E = 2*vdwtype[jnrE+0];
911 vdwjidx0F = 2*vdwtype[jnrF+0];
912 vdwjidx0G = 2*vdwtype[jnrG+0];
913 vdwjidx0H = 2*vdwtype[jnrH+0];
915 /**************************
916 * CALCULATE INTERACTIONS *
917 **************************/
919 r00 = _mm256_mul_ps(rsq00,rinv00);
920 r00 = _mm256_andnot_ps(dummy_mask,r00);
922 /* Compute parameters for interactions between i and j atoms */
923 qq00 = _mm256_mul_ps(iq0,jq0);
924 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
925 vdwioffsetptr0+vdwjidx0B,
926 vdwioffsetptr0+vdwjidx0C,
927 vdwioffsetptr0+vdwjidx0D,
928 vdwioffsetptr0+vdwjidx0E,
929 vdwioffsetptr0+vdwjidx0F,
930 vdwioffsetptr0+vdwjidx0G,
931 vdwioffsetptr0+vdwjidx0H,
934 /* Calculate table index by multiplying r with table scale and truncate to integer */
935 rt = _mm256_mul_ps(r00,vftabscale);
936 vfitab = _mm256_cvttps_epi32(rt);
937 vfeps = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
938 /* AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
939 vfitab_lo = _mm256_extractf128_si256(vfitab,0x0);
940 vfitab_hi = _mm256_extractf128_si256(vfitab,0x1);
941 vfitab_lo = _mm_slli_epi32(vfitab_lo,3);
942 vfitab_hi = _mm_slli_epi32(vfitab_hi,3);
944 /* EWALD ELECTROSTATICS */
946 /* Analytical PME correction */
947 zeta2 = _mm256_mul_ps(beta2,rsq00);
948 rinv3 = _mm256_mul_ps(rinvsq00,rinv00);
949 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
950 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
951 felec = _mm256_mul_ps(qq00,felec);
953 /* CUBIC SPLINE TABLE DISPERSION */
954 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
955 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
956 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
957 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
958 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
959 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
960 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
961 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
962 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
963 Heps = _mm256_mul_ps(vfeps,H);
964 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
965 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
966 fvdw6 = _mm256_mul_ps(c6_00,FF);
968 /* CUBIC SPLINE TABLE REPULSION */
969 vfitab_lo = _mm_add_epi32(vfitab_lo,ifour);
970 vfitab_hi = _mm_add_epi32(vfitab_hi,ifour);
971 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
972 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
973 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
974 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
975 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
976 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
977 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
978 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
979 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
980 Heps = _mm256_mul_ps(vfeps,H);
981 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
982 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
983 fvdw12 = _mm256_mul_ps(c12_00,FF);
984 fvdw = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
986 fscal = _mm256_add_ps(felec,fvdw);
988 fscal = _mm256_andnot_ps(dummy_mask,fscal);
990 /* Calculate temporary vectorial force */
991 tx = _mm256_mul_ps(fscal,dx00);
992 ty = _mm256_mul_ps(fscal,dy00);
993 tz = _mm256_mul_ps(fscal,dz00);
995 /* Update vectorial force */
996 fix0 = _mm256_add_ps(fix0,tx);
997 fiy0 = _mm256_add_ps(fiy0,ty);
998 fiz0 = _mm256_add_ps(fiz0,tz);
1000 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1001 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1002 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1003 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1004 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1005 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1006 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1007 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1008 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
1010 /* Inner loop uses 83 flops */
1013 /* End of innermost loop */
1015 gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
1016 f+i_coord_offset,fshift+i_shift_offset);
1018 /* Increment number of inner iterations */
1019 inneriter += j_index_end - j_index_start;
1021 /* Outer loop uses 7 flops */
1024 /* Increment number of outer iterations */
1027 /* Update outer/inner flops */
1029 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*83);