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_ElecRF_VdwCSTab_GeomP1P1_VF_avx_256_single
54 * Electrostatics interaction: ReactionField
55 * VdW interaction: CubicSplineTable
56 * Geometry: Particle-Particle
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecRF_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;
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;
123 krf = _mm256_set1_ps(fr->ic->k_rf);
124 krf2 = _mm256_set1_ps(fr->ic->k_rf*2.0);
125 crf = _mm256_set1_ps(fr->ic->c_rf);
126 nvdwtype = fr->ntype;
128 vdwtype = mdatoms->typeA;
130 vftab = kernel_data->table_vdw->data;
131 vftabscale = _mm256_set1_ps(kernel_data->table_vdw->scale);
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_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
169 fix0 = _mm256_setzero_ps();
170 fiy0 = _mm256_setzero_ps();
171 fiz0 = _mm256_setzero_ps();
173 /* Load parameters for i particles */
174 iq0 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
175 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
177 /* Reset potential sums */
178 velecsum = _mm256_setzero_ps();
179 vvdwsum = _mm256_setzero_ps();
181 /* Start inner kernel loop */
182 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
185 /* Get j neighbor index, and coordinate index */
194 j_coord_offsetA = DIM*jnrA;
195 j_coord_offsetB = DIM*jnrB;
196 j_coord_offsetC = DIM*jnrC;
197 j_coord_offsetD = DIM*jnrD;
198 j_coord_offsetE = DIM*jnrE;
199 j_coord_offsetF = DIM*jnrF;
200 j_coord_offsetG = DIM*jnrG;
201 j_coord_offsetH = DIM*jnrH;
203 /* load j atom coordinates */
204 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
205 x+j_coord_offsetC,x+j_coord_offsetD,
206 x+j_coord_offsetE,x+j_coord_offsetF,
207 x+j_coord_offsetG,x+j_coord_offsetH,
210 /* Calculate displacement vector */
211 dx00 = _mm256_sub_ps(ix0,jx0);
212 dy00 = _mm256_sub_ps(iy0,jy0);
213 dz00 = _mm256_sub_ps(iz0,jz0);
215 /* Calculate squared distance and things based on it */
216 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
218 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
220 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
222 /* Load parameters for j particles */
223 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
224 charge+jnrC+0,charge+jnrD+0,
225 charge+jnrE+0,charge+jnrF+0,
226 charge+jnrG+0,charge+jnrH+0);
227 vdwjidx0A = 2*vdwtype[jnrA+0];
228 vdwjidx0B = 2*vdwtype[jnrB+0];
229 vdwjidx0C = 2*vdwtype[jnrC+0];
230 vdwjidx0D = 2*vdwtype[jnrD+0];
231 vdwjidx0E = 2*vdwtype[jnrE+0];
232 vdwjidx0F = 2*vdwtype[jnrF+0];
233 vdwjidx0G = 2*vdwtype[jnrG+0];
234 vdwjidx0H = 2*vdwtype[jnrH+0];
236 /**************************
237 * CALCULATE INTERACTIONS *
238 **************************/
240 r00 = _mm256_mul_ps(rsq00,rinv00);
242 /* Compute parameters for interactions between i and j atoms */
243 qq00 = _mm256_mul_ps(iq0,jq0);
244 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
245 vdwioffsetptr0+vdwjidx0B,
246 vdwioffsetptr0+vdwjidx0C,
247 vdwioffsetptr0+vdwjidx0D,
248 vdwioffsetptr0+vdwjidx0E,
249 vdwioffsetptr0+vdwjidx0F,
250 vdwioffsetptr0+vdwjidx0G,
251 vdwioffsetptr0+vdwjidx0H,
254 /* Calculate table index by multiplying r with table scale and truncate to integer */
255 rt = _mm256_mul_ps(r00,vftabscale);
256 vfitab = _mm256_cvttps_epi32(rt);
257 vfeps = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
258 /* AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
259 vfitab_lo = _mm256_extractf128_si256(vfitab,0x0);
260 vfitab_hi = _mm256_extractf128_si256(vfitab,0x1);
261 vfitab_lo = _mm_slli_epi32(vfitab_lo,3);
262 vfitab_hi = _mm_slli_epi32(vfitab_hi,3);
264 /* REACTION-FIELD ELECTROSTATICS */
265 velec = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
266 felec = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
268 /* CUBIC SPLINE TABLE DISPERSION */
269 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
270 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
271 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
272 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
273 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
274 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
275 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
276 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
277 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
278 Heps = _mm256_mul_ps(vfeps,H);
279 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
280 VV = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
281 vvdw6 = _mm256_mul_ps(c6_00,VV);
282 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
283 fvdw6 = _mm256_mul_ps(c6_00,FF);
285 /* CUBIC SPLINE TABLE REPULSION */
286 vfitab_lo = _mm_add_epi32(vfitab_lo,ifour);
287 vfitab_hi = _mm_add_epi32(vfitab_hi,ifour);
288 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
289 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
290 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
291 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
292 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
293 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
294 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
295 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
296 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
297 Heps = _mm256_mul_ps(vfeps,H);
298 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
299 VV = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
300 vvdw12 = _mm256_mul_ps(c12_00,VV);
301 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
302 fvdw12 = _mm256_mul_ps(c12_00,FF);
303 vvdw = _mm256_add_ps(vvdw12,vvdw6);
304 fvdw = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
306 /* Update potential sum for this i atom from the interaction with this j atom. */
307 velecsum = _mm256_add_ps(velecsum,velec);
308 vvdwsum = _mm256_add_ps(vvdwsum,vvdw);
310 fscal = _mm256_add_ps(felec,fvdw);
312 /* Calculate temporary vectorial force */
313 tx = _mm256_mul_ps(fscal,dx00);
314 ty = _mm256_mul_ps(fscal,dy00);
315 tz = _mm256_mul_ps(fscal,dz00);
317 /* Update vectorial force */
318 fix0 = _mm256_add_ps(fix0,tx);
319 fiy0 = _mm256_add_ps(fiy0,ty);
320 fiz0 = _mm256_add_ps(fiz0,tz);
322 fjptrA = f+j_coord_offsetA;
323 fjptrB = f+j_coord_offsetB;
324 fjptrC = f+j_coord_offsetC;
325 fjptrD = f+j_coord_offsetD;
326 fjptrE = f+j_coord_offsetE;
327 fjptrF = f+j_coord_offsetF;
328 fjptrG = f+j_coord_offsetG;
329 fjptrH = f+j_coord_offsetH;
330 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
332 /* Inner loop uses 67 flops */
338 /* Get j neighbor index, and coordinate index */
339 jnrlistA = jjnr[jidx];
340 jnrlistB = jjnr[jidx+1];
341 jnrlistC = jjnr[jidx+2];
342 jnrlistD = jjnr[jidx+3];
343 jnrlistE = jjnr[jidx+4];
344 jnrlistF = jjnr[jidx+5];
345 jnrlistG = jjnr[jidx+6];
346 jnrlistH = jjnr[jidx+7];
347 /* Sign of each element will be negative for non-real atoms.
348 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
349 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
351 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
352 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
354 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
355 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
356 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
357 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
358 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
359 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
360 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
361 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
362 j_coord_offsetA = DIM*jnrA;
363 j_coord_offsetB = DIM*jnrB;
364 j_coord_offsetC = DIM*jnrC;
365 j_coord_offsetD = DIM*jnrD;
366 j_coord_offsetE = DIM*jnrE;
367 j_coord_offsetF = DIM*jnrF;
368 j_coord_offsetG = DIM*jnrG;
369 j_coord_offsetH = DIM*jnrH;
371 /* load j atom coordinates */
372 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
373 x+j_coord_offsetC,x+j_coord_offsetD,
374 x+j_coord_offsetE,x+j_coord_offsetF,
375 x+j_coord_offsetG,x+j_coord_offsetH,
378 /* Calculate displacement vector */
379 dx00 = _mm256_sub_ps(ix0,jx0);
380 dy00 = _mm256_sub_ps(iy0,jy0);
381 dz00 = _mm256_sub_ps(iz0,jz0);
383 /* Calculate squared distance and things based on it */
384 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
386 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
388 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
390 /* Load parameters for j particles */
391 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
392 charge+jnrC+0,charge+jnrD+0,
393 charge+jnrE+0,charge+jnrF+0,
394 charge+jnrG+0,charge+jnrH+0);
395 vdwjidx0A = 2*vdwtype[jnrA+0];
396 vdwjidx0B = 2*vdwtype[jnrB+0];
397 vdwjidx0C = 2*vdwtype[jnrC+0];
398 vdwjidx0D = 2*vdwtype[jnrD+0];
399 vdwjidx0E = 2*vdwtype[jnrE+0];
400 vdwjidx0F = 2*vdwtype[jnrF+0];
401 vdwjidx0G = 2*vdwtype[jnrG+0];
402 vdwjidx0H = 2*vdwtype[jnrH+0];
404 /**************************
405 * CALCULATE INTERACTIONS *
406 **************************/
408 r00 = _mm256_mul_ps(rsq00,rinv00);
409 r00 = _mm256_andnot_ps(dummy_mask,r00);
411 /* Compute parameters for interactions between i and j atoms */
412 qq00 = _mm256_mul_ps(iq0,jq0);
413 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
414 vdwioffsetptr0+vdwjidx0B,
415 vdwioffsetptr0+vdwjidx0C,
416 vdwioffsetptr0+vdwjidx0D,
417 vdwioffsetptr0+vdwjidx0E,
418 vdwioffsetptr0+vdwjidx0F,
419 vdwioffsetptr0+vdwjidx0G,
420 vdwioffsetptr0+vdwjidx0H,
423 /* Calculate table index by multiplying r with table scale and truncate to integer */
424 rt = _mm256_mul_ps(r00,vftabscale);
425 vfitab = _mm256_cvttps_epi32(rt);
426 vfeps = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
427 /* AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
428 vfitab_lo = _mm256_extractf128_si256(vfitab,0x0);
429 vfitab_hi = _mm256_extractf128_si256(vfitab,0x1);
430 vfitab_lo = _mm_slli_epi32(vfitab_lo,3);
431 vfitab_hi = _mm_slli_epi32(vfitab_hi,3);
433 /* REACTION-FIELD ELECTROSTATICS */
434 velec = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
435 felec = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
437 /* CUBIC SPLINE TABLE DISPERSION */
438 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
439 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
440 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
441 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
442 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
443 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
444 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
445 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
446 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
447 Heps = _mm256_mul_ps(vfeps,H);
448 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
449 VV = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
450 vvdw6 = _mm256_mul_ps(c6_00,VV);
451 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
452 fvdw6 = _mm256_mul_ps(c6_00,FF);
454 /* CUBIC SPLINE TABLE REPULSION */
455 vfitab_lo = _mm_add_epi32(vfitab_lo,ifour);
456 vfitab_hi = _mm_add_epi32(vfitab_hi,ifour);
457 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
458 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
459 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
460 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
461 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
462 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
463 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
464 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
465 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
466 Heps = _mm256_mul_ps(vfeps,H);
467 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
468 VV = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
469 vvdw12 = _mm256_mul_ps(c12_00,VV);
470 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
471 fvdw12 = _mm256_mul_ps(c12_00,FF);
472 vvdw = _mm256_add_ps(vvdw12,vvdw6);
473 fvdw = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
475 /* Update potential sum for this i atom from the interaction with this j atom. */
476 velec = _mm256_andnot_ps(dummy_mask,velec);
477 velecsum = _mm256_add_ps(velecsum,velec);
478 vvdw = _mm256_andnot_ps(dummy_mask,vvdw);
479 vvdwsum = _mm256_add_ps(vvdwsum,vvdw);
481 fscal = _mm256_add_ps(felec,fvdw);
483 fscal = _mm256_andnot_ps(dummy_mask,fscal);
485 /* Calculate temporary vectorial force */
486 tx = _mm256_mul_ps(fscal,dx00);
487 ty = _mm256_mul_ps(fscal,dy00);
488 tz = _mm256_mul_ps(fscal,dz00);
490 /* Update vectorial force */
491 fix0 = _mm256_add_ps(fix0,tx);
492 fiy0 = _mm256_add_ps(fiy0,ty);
493 fiz0 = _mm256_add_ps(fiz0,tz);
495 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
496 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
497 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
498 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
499 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
500 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
501 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
502 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
503 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
505 /* Inner loop uses 68 flops */
508 /* End of innermost loop */
510 gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
511 f+i_coord_offset,fshift+i_shift_offset);
514 /* Update potential energies */
515 gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
516 gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
518 /* Increment number of inner iterations */
519 inneriter += j_index_end - j_index_start;
521 /* Outer loop uses 9 flops */
524 /* Increment number of outer iterations */
527 /* Update outer/inner flops */
529 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*68);
532 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwCSTab_GeomP1P1_F_avx_256_single
533 * Electrostatics interaction: ReactionField
534 * VdW interaction: CubicSplineTable
535 * Geometry: Particle-Particle
536 * Calculate force/pot: Force
539 nb_kernel_ElecRF_VdwCSTab_GeomP1P1_F_avx_256_single
540 (t_nblist * gmx_restrict nlist,
541 rvec * gmx_restrict xx,
542 rvec * gmx_restrict ff,
543 t_forcerec * gmx_restrict fr,
544 t_mdatoms * gmx_restrict mdatoms,
545 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
546 t_nrnb * gmx_restrict nrnb)
548 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
549 * just 0 for non-waters.
550 * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
551 * jnr indices corresponding to data put in the four positions in the SIMD register.
553 int i_shift_offset,i_coord_offset,outeriter,inneriter;
554 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
555 int jnrA,jnrB,jnrC,jnrD;
556 int jnrE,jnrF,jnrG,jnrH;
557 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
558 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
559 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
560 int j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
561 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
563 real *shiftvec,*fshift,*x,*f;
564 real *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
566 __m256 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
567 real * vdwioffsetptr0;
568 __m256 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
569 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
570 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
571 __m256 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
572 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
575 __m256 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
578 __m256 one_sixth = _mm256_set1_ps(1.0/6.0);
579 __m256 one_twelfth = _mm256_set1_ps(1.0/12.0);
581 __m128i vfitab_lo,vfitab_hi;
582 __m128i ifour = _mm_set1_epi32(4);
583 __m256 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
585 __m256 dummy_mask,cutoff_mask;
586 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
587 __m256 one = _mm256_set1_ps(1.0);
588 __m256 two = _mm256_set1_ps(2.0);
594 jindex = nlist->jindex;
596 shiftidx = nlist->shift;
598 shiftvec = fr->shift_vec[0];
599 fshift = fr->fshift[0];
600 facel = _mm256_set1_ps(fr->epsfac);
601 charge = mdatoms->chargeA;
602 krf = _mm256_set1_ps(fr->ic->k_rf);
603 krf2 = _mm256_set1_ps(fr->ic->k_rf*2.0);
604 crf = _mm256_set1_ps(fr->ic->c_rf);
605 nvdwtype = fr->ntype;
607 vdwtype = mdatoms->typeA;
609 vftab = kernel_data->table_vdw->data;
610 vftabscale = _mm256_set1_ps(kernel_data->table_vdw->scale);
612 /* Avoid stupid compiler warnings */
613 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
626 for(iidx=0;iidx<4*DIM;iidx++)
631 /* Start outer loop over neighborlists */
632 for(iidx=0; iidx<nri; iidx++)
634 /* Load shift vector for this list */
635 i_shift_offset = DIM*shiftidx[iidx];
637 /* Load limits for loop over neighbors */
638 j_index_start = jindex[iidx];
639 j_index_end = jindex[iidx+1];
641 /* Get outer coordinate index */
643 i_coord_offset = DIM*inr;
645 /* Load i particle coords and add shift vector */
646 gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
648 fix0 = _mm256_setzero_ps();
649 fiy0 = _mm256_setzero_ps();
650 fiz0 = _mm256_setzero_ps();
652 /* Load parameters for i particles */
653 iq0 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
654 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
656 /* Start inner kernel loop */
657 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
660 /* Get j neighbor index, and coordinate index */
669 j_coord_offsetA = DIM*jnrA;
670 j_coord_offsetB = DIM*jnrB;
671 j_coord_offsetC = DIM*jnrC;
672 j_coord_offsetD = DIM*jnrD;
673 j_coord_offsetE = DIM*jnrE;
674 j_coord_offsetF = DIM*jnrF;
675 j_coord_offsetG = DIM*jnrG;
676 j_coord_offsetH = DIM*jnrH;
678 /* load j atom coordinates */
679 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
680 x+j_coord_offsetC,x+j_coord_offsetD,
681 x+j_coord_offsetE,x+j_coord_offsetF,
682 x+j_coord_offsetG,x+j_coord_offsetH,
685 /* Calculate displacement vector */
686 dx00 = _mm256_sub_ps(ix0,jx0);
687 dy00 = _mm256_sub_ps(iy0,jy0);
688 dz00 = _mm256_sub_ps(iz0,jz0);
690 /* Calculate squared distance and things based on it */
691 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
693 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
695 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
697 /* Load parameters for j particles */
698 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
699 charge+jnrC+0,charge+jnrD+0,
700 charge+jnrE+0,charge+jnrF+0,
701 charge+jnrG+0,charge+jnrH+0);
702 vdwjidx0A = 2*vdwtype[jnrA+0];
703 vdwjidx0B = 2*vdwtype[jnrB+0];
704 vdwjidx0C = 2*vdwtype[jnrC+0];
705 vdwjidx0D = 2*vdwtype[jnrD+0];
706 vdwjidx0E = 2*vdwtype[jnrE+0];
707 vdwjidx0F = 2*vdwtype[jnrF+0];
708 vdwjidx0G = 2*vdwtype[jnrG+0];
709 vdwjidx0H = 2*vdwtype[jnrH+0];
711 /**************************
712 * CALCULATE INTERACTIONS *
713 **************************/
715 r00 = _mm256_mul_ps(rsq00,rinv00);
717 /* Compute parameters for interactions between i and j atoms */
718 qq00 = _mm256_mul_ps(iq0,jq0);
719 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
720 vdwioffsetptr0+vdwjidx0B,
721 vdwioffsetptr0+vdwjidx0C,
722 vdwioffsetptr0+vdwjidx0D,
723 vdwioffsetptr0+vdwjidx0E,
724 vdwioffsetptr0+vdwjidx0F,
725 vdwioffsetptr0+vdwjidx0G,
726 vdwioffsetptr0+vdwjidx0H,
729 /* Calculate table index by multiplying r with table scale and truncate to integer */
730 rt = _mm256_mul_ps(r00,vftabscale);
731 vfitab = _mm256_cvttps_epi32(rt);
732 vfeps = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
733 /* AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
734 vfitab_lo = _mm256_extractf128_si256(vfitab,0x0);
735 vfitab_hi = _mm256_extractf128_si256(vfitab,0x1);
736 vfitab_lo = _mm_slli_epi32(vfitab_lo,3);
737 vfitab_hi = _mm_slli_epi32(vfitab_hi,3);
739 /* REACTION-FIELD ELECTROSTATICS */
740 felec = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
742 /* CUBIC SPLINE TABLE DISPERSION */
743 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
744 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
745 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
746 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
747 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
748 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
749 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
750 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
751 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
752 Heps = _mm256_mul_ps(vfeps,H);
753 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
754 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
755 fvdw6 = _mm256_mul_ps(c6_00,FF);
757 /* CUBIC SPLINE TABLE REPULSION */
758 vfitab_lo = _mm_add_epi32(vfitab_lo,ifour);
759 vfitab_hi = _mm_add_epi32(vfitab_hi,ifour);
760 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
761 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
762 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
763 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
764 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
765 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
766 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
767 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
768 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
769 Heps = _mm256_mul_ps(vfeps,H);
770 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
771 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
772 fvdw12 = _mm256_mul_ps(c12_00,FF);
773 fvdw = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
775 fscal = _mm256_add_ps(felec,fvdw);
777 /* Calculate temporary vectorial force */
778 tx = _mm256_mul_ps(fscal,dx00);
779 ty = _mm256_mul_ps(fscal,dy00);
780 tz = _mm256_mul_ps(fscal,dz00);
782 /* Update vectorial force */
783 fix0 = _mm256_add_ps(fix0,tx);
784 fiy0 = _mm256_add_ps(fiy0,ty);
785 fiz0 = _mm256_add_ps(fiz0,tz);
787 fjptrA = f+j_coord_offsetA;
788 fjptrB = f+j_coord_offsetB;
789 fjptrC = f+j_coord_offsetC;
790 fjptrD = f+j_coord_offsetD;
791 fjptrE = f+j_coord_offsetE;
792 fjptrF = f+j_coord_offsetF;
793 fjptrG = f+j_coord_offsetG;
794 fjptrH = f+j_coord_offsetH;
795 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
797 /* Inner loop uses 54 flops */
803 /* Get j neighbor index, and coordinate index */
804 jnrlistA = jjnr[jidx];
805 jnrlistB = jjnr[jidx+1];
806 jnrlistC = jjnr[jidx+2];
807 jnrlistD = jjnr[jidx+3];
808 jnrlistE = jjnr[jidx+4];
809 jnrlistF = jjnr[jidx+5];
810 jnrlistG = jjnr[jidx+6];
811 jnrlistH = jjnr[jidx+7];
812 /* Sign of each element will be negative for non-real atoms.
813 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
814 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
816 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
817 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
819 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
820 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
821 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
822 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
823 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
824 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
825 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
826 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
827 j_coord_offsetA = DIM*jnrA;
828 j_coord_offsetB = DIM*jnrB;
829 j_coord_offsetC = DIM*jnrC;
830 j_coord_offsetD = DIM*jnrD;
831 j_coord_offsetE = DIM*jnrE;
832 j_coord_offsetF = DIM*jnrF;
833 j_coord_offsetG = DIM*jnrG;
834 j_coord_offsetH = DIM*jnrH;
836 /* load j atom coordinates */
837 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
838 x+j_coord_offsetC,x+j_coord_offsetD,
839 x+j_coord_offsetE,x+j_coord_offsetF,
840 x+j_coord_offsetG,x+j_coord_offsetH,
843 /* Calculate displacement vector */
844 dx00 = _mm256_sub_ps(ix0,jx0);
845 dy00 = _mm256_sub_ps(iy0,jy0);
846 dz00 = _mm256_sub_ps(iz0,jz0);
848 /* Calculate squared distance and things based on it */
849 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
851 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
853 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
855 /* Load parameters for j particles */
856 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
857 charge+jnrC+0,charge+jnrD+0,
858 charge+jnrE+0,charge+jnrF+0,
859 charge+jnrG+0,charge+jnrH+0);
860 vdwjidx0A = 2*vdwtype[jnrA+0];
861 vdwjidx0B = 2*vdwtype[jnrB+0];
862 vdwjidx0C = 2*vdwtype[jnrC+0];
863 vdwjidx0D = 2*vdwtype[jnrD+0];
864 vdwjidx0E = 2*vdwtype[jnrE+0];
865 vdwjidx0F = 2*vdwtype[jnrF+0];
866 vdwjidx0G = 2*vdwtype[jnrG+0];
867 vdwjidx0H = 2*vdwtype[jnrH+0];
869 /**************************
870 * CALCULATE INTERACTIONS *
871 **************************/
873 r00 = _mm256_mul_ps(rsq00,rinv00);
874 r00 = _mm256_andnot_ps(dummy_mask,r00);
876 /* Compute parameters for interactions between i and j atoms */
877 qq00 = _mm256_mul_ps(iq0,jq0);
878 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
879 vdwioffsetptr0+vdwjidx0B,
880 vdwioffsetptr0+vdwjidx0C,
881 vdwioffsetptr0+vdwjidx0D,
882 vdwioffsetptr0+vdwjidx0E,
883 vdwioffsetptr0+vdwjidx0F,
884 vdwioffsetptr0+vdwjidx0G,
885 vdwioffsetptr0+vdwjidx0H,
888 /* Calculate table index by multiplying r with table scale and truncate to integer */
889 rt = _mm256_mul_ps(r00,vftabscale);
890 vfitab = _mm256_cvttps_epi32(rt);
891 vfeps = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
892 /* AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
893 vfitab_lo = _mm256_extractf128_si256(vfitab,0x0);
894 vfitab_hi = _mm256_extractf128_si256(vfitab,0x1);
895 vfitab_lo = _mm_slli_epi32(vfitab_lo,3);
896 vfitab_hi = _mm_slli_epi32(vfitab_hi,3);
898 /* REACTION-FIELD ELECTROSTATICS */
899 felec = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
901 /* CUBIC SPLINE TABLE DISPERSION */
902 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
903 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
904 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
905 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
906 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
907 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
908 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
909 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
910 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
911 Heps = _mm256_mul_ps(vfeps,H);
912 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
913 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
914 fvdw6 = _mm256_mul_ps(c6_00,FF);
916 /* CUBIC SPLINE TABLE REPULSION */
917 vfitab_lo = _mm_add_epi32(vfitab_lo,ifour);
918 vfitab_hi = _mm_add_epi32(vfitab_hi,ifour);
919 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
920 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
921 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
922 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
923 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
924 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
925 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
926 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
927 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
928 Heps = _mm256_mul_ps(vfeps,H);
929 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
930 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
931 fvdw12 = _mm256_mul_ps(c12_00,FF);
932 fvdw = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
934 fscal = _mm256_add_ps(felec,fvdw);
936 fscal = _mm256_andnot_ps(dummy_mask,fscal);
938 /* Calculate temporary vectorial force */
939 tx = _mm256_mul_ps(fscal,dx00);
940 ty = _mm256_mul_ps(fscal,dy00);
941 tz = _mm256_mul_ps(fscal,dz00);
943 /* Update vectorial force */
944 fix0 = _mm256_add_ps(fix0,tx);
945 fiy0 = _mm256_add_ps(fiy0,ty);
946 fiz0 = _mm256_add_ps(fiz0,tz);
948 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
949 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
950 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
951 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
952 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
953 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
954 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
955 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
956 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
958 /* Inner loop uses 55 flops */
961 /* End of innermost loop */
963 gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
964 f+i_coord_offset,fshift+i_shift_offset);
966 /* Increment number of inner iterations */
967 inneriter += j_index_end - j_index_start;
969 /* Outer loop uses 7 flops */
972 /* Increment number of outer iterations */
975 /* Update outer/inner flops */
977 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*55);