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_VdwCSTab_GeomP1P1_VF_avx_256_single
54 * Electrostatics interaction: CubicSplineTable
55 * VdW interaction: CubicSplineTable
56 * Geometry: Particle-Particle
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecCSTab_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 nvdwtype = fr->ntype;
125 vdwtype = mdatoms->typeA;
127 vftab = kernel_data->table_elec_vdw->data;
128 vftabscale = _mm256_set1_ps(kernel_data->table_elec_vdw->scale);
130 /* Avoid stupid compiler warnings */
131 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
144 for(iidx=0;iidx<4*DIM;iidx++)
149 /* Start outer loop over neighborlists */
150 for(iidx=0; iidx<nri; iidx++)
152 /* Load shift vector for this list */
153 i_shift_offset = DIM*shiftidx[iidx];
155 /* Load limits for loop over neighbors */
156 j_index_start = jindex[iidx];
157 j_index_end = jindex[iidx+1];
159 /* Get outer coordinate index */
161 i_coord_offset = DIM*inr;
163 /* Load i particle coords and add shift vector */
164 gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
166 fix0 = _mm256_setzero_ps();
167 fiy0 = _mm256_setzero_ps();
168 fiz0 = _mm256_setzero_ps();
170 /* Load parameters for i particles */
171 iq0 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
172 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
174 /* Reset potential sums */
175 velecsum = _mm256_setzero_ps();
176 vvdwsum = _mm256_setzero_ps();
178 /* Start inner kernel loop */
179 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
182 /* Get j neighbor index, and coordinate index */
191 j_coord_offsetA = DIM*jnrA;
192 j_coord_offsetB = DIM*jnrB;
193 j_coord_offsetC = DIM*jnrC;
194 j_coord_offsetD = DIM*jnrD;
195 j_coord_offsetE = DIM*jnrE;
196 j_coord_offsetF = DIM*jnrF;
197 j_coord_offsetG = DIM*jnrG;
198 j_coord_offsetH = DIM*jnrH;
200 /* load j atom coordinates */
201 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
202 x+j_coord_offsetC,x+j_coord_offsetD,
203 x+j_coord_offsetE,x+j_coord_offsetF,
204 x+j_coord_offsetG,x+j_coord_offsetH,
207 /* Calculate displacement vector */
208 dx00 = _mm256_sub_ps(ix0,jx0);
209 dy00 = _mm256_sub_ps(iy0,jy0);
210 dz00 = _mm256_sub_ps(iz0,jz0);
212 /* Calculate squared distance and things based on it */
213 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
215 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
217 /* Load parameters for j particles */
218 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
219 charge+jnrC+0,charge+jnrD+0,
220 charge+jnrE+0,charge+jnrF+0,
221 charge+jnrG+0,charge+jnrH+0);
222 vdwjidx0A = 2*vdwtype[jnrA+0];
223 vdwjidx0B = 2*vdwtype[jnrB+0];
224 vdwjidx0C = 2*vdwtype[jnrC+0];
225 vdwjidx0D = 2*vdwtype[jnrD+0];
226 vdwjidx0E = 2*vdwtype[jnrE+0];
227 vdwjidx0F = 2*vdwtype[jnrF+0];
228 vdwjidx0G = 2*vdwtype[jnrG+0];
229 vdwjidx0H = 2*vdwtype[jnrH+0];
231 /**************************
232 * CALCULATE INTERACTIONS *
233 **************************/
235 r00 = _mm256_mul_ps(rsq00,rinv00);
237 /* Compute parameters for interactions between i and j atoms */
238 qq00 = _mm256_mul_ps(iq0,jq0);
239 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
240 vdwioffsetptr0+vdwjidx0B,
241 vdwioffsetptr0+vdwjidx0C,
242 vdwioffsetptr0+vdwjidx0D,
243 vdwioffsetptr0+vdwjidx0E,
244 vdwioffsetptr0+vdwjidx0F,
245 vdwioffsetptr0+vdwjidx0G,
246 vdwioffsetptr0+vdwjidx0H,
249 /* Calculate table index by multiplying r with table scale and truncate to integer */
250 rt = _mm256_mul_ps(r00,vftabscale);
251 vfitab = _mm256_cvttps_epi32(rt);
252 vfeps = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
253 /* AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
254 vfitab_lo = _mm256_extractf128_si256(vfitab,0x0);
255 vfitab_hi = _mm256_extractf128_si256(vfitab,0x1);
256 vfitab_lo = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
257 vfitab_hi = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
259 /* CUBIC SPLINE TABLE ELECTROSTATICS */
260 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
261 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
262 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
263 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
264 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
265 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
266 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
267 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
268 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
269 Heps = _mm256_mul_ps(vfeps,H);
270 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
271 VV = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
272 velec = _mm256_mul_ps(qq00,VV);
273 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
274 felec = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
276 /* CUBIC SPLINE TABLE DISPERSION */
277 vfitab_lo = _mm_add_epi32(vfitab_lo,ifour);
278 vfitab_hi = _mm_add_epi32(vfitab_hi,ifour);
279 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
280 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
281 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
282 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
283 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
284 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
285 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
286 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
287 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
288 Heps = _mm256_mul_ps(vfeps,H);
289 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
290 VV = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
291 vvdw6 = _mm256_mul_ps(c6_00,VV);
292 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
293 fvdw6 = _mm256_mul_ps(c6_00,FF);
295 /* CUBIC SPLINE TABLE REPULSION */
296 vfitab_lo = _mm_add_epi32(vfitab_lo,ifour);
297 vfitab_hi = _mm_add_epi32(vfitab_hi,ifour);
298 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
299 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
300 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
301 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
302 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
303 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
304 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
305 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
306 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
307 Heps = _mm256_mul_ps(vfeps,H);
308 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
309 VV = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
310 vvdw12 = _mm256_mul_ps(c12_00,VV);
311 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
312 fvdw12 = _mm256_mul_ps(c12_00,FF);
313 vvdw = _mm256_add_ps(vvdw12,vvdw6);
314 fvdw = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
316 /* Update potential sum for this i atom from the interaction with this j atom. */
317 velecsum = _mm256_add_ps(velecsum,velec);
318 vvdwsum = _mm256_add_ps(vvdwsum,vvdw);
320 fscal = _mm256_add_ps(felec,fvdw);
322 /* Calculate temporary vectorial force */
323 tx = _mm256_mul_ps(fscal,dx00);
324 ty = _mm256_mul_ps(fscal,dy00);
325 tz = _mm256_mul_ps(fscal,dz00);
327 /* Update vectorial force */
328 fix0 = _mm256_add_ps(fix0,tx);
329 fiy0 = _mm256_add_ps(fiy0,ty);
330 fiz0 = _mm256_add_ps(fiz0,tz);
332 fjptrA = f+j_coord_offsetA;
333 fjptrB = f+j_coord_offsetB;
334 fjptrC = f+j_coord_offsetC;
335 fjptrD = f+j_coord_offsetD;
336 fjptrE = f+j_coord_offsetE;
337 fjptrF = f+j_coord_offsetF;
338 fjptrG = f+j_coord_offsetG;
339 fjptrH = f+j_coord_offsetH;
340 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
342 /* Inner loop uses 73 flops */
348 /* Get j neighbor index, and coordinate index */
349 jnrlistA = jjnr[jidx];
350 jnrlistB = jjnr[jidx+1];
351 jnrlistC = jjnr[jidx+2];
352 jnrlistD = jjnr[jidx+3];
353 jnrlistE = jjnr[jidx+4];
354 jnrlistF = jjnr[jidx+5];
355 jnrlistG = jjnr[jidx+6];
356 jnrlistH = jjnr[jidx+7];
357 /* Sign of each element will be negative for non-real atoms.
358 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
359 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
361 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
362 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
364 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
365 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
366 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
367 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
368 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
369 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
370 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
371 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
372 j_coord_offsetA = DIM*jnrA;
373 j_coord_offsetB = DIM*jnrB;
374 j_coord_offsetC = DIM*jnrC;
375 j_coord_offsetD = DIM*jnrD;
376 j_coord_offsetE = DIM*jnrE;
377 j_coord_offsetF = DIM*jnrF;
378 j_coord_offsetG = DIM*jnrG;
379 j_coord_offsetH = DIM*jnrH;
381 /* load j atom coordinates */
382 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
383 x+j_coord_offsetC,x+j_coord_offsetD,
384 x+j_coord_offsetE,x+j_coord_offsetF,
385 x+j_coord_offsetG,x+j_coord_offsetH,
388 /* Calculate displacement vector */
389 dx00 = _mm256_sub_ps(ix0,jx0);
390 dy00 = _mm256_sub_ps(iy0,jy0);
391 dz00 = _mm256_sub_ps(iz0,jz0);
393 /* Calculate squared distance and things based on it */
394 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
396 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
398 /* Load parameters for j particles */
399 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
400 charge+jnrC+0,charge+jnrD+0,
401 charge+jnrE+0,charge+jnrF+0,
402 charge+jnrG+0,charge+jnrH+0);
403 vdwjidx0A = 2*vdwtype[jnrA+0];
404 vdwjidx0B = 2*vdwtype[jnrB+0];
405 vdwjidx0C = 2*vdwtype[jnrC+0];
406 vdwjidx0D = 2*vdwtype[jnrD+0];
407 vdwjidx0E = 2*vdwtype[jnrE+0];
408 vdwjidx0F = 2*vdwtype[jnrF+0];
409 vdwjidx0G = 2*vdwtype[jnrG+0];
410 vdwjidx0H = 2*vdwtype[jnrH+0];
412 /**************************
413 * CALCULATE INTERACTIONS *
414 **************************/
416 r00 = _mm256_mul_ps(rsq00,rinv00);
417 r00 = _mm256_andnot_ps(dummy_mask,r00);
419 /* Compute parameters for interactions between i and j atoms */
420 qq00 = _mm256_mul_ps(iq0,jq0);
421 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
422 vdwioffsetptr0+vdwjidx0B,
423 vdwioffsetptr0+vdwjidx0C,
424 vdwioffsetptr0+vdwjidx0D,
425 vdwioffsetptr0+vdwjidx0E,
426 vdwioffsetptr0+vdwjidx0F,
427 vdwioffsetptr0+vdwjidx0G,
428 vdwioffsetptr0+vdwjidx0H,
431 /* Calculate table index by multiplying r with table scale and truncate to integer */
432 rt = _mm256_mul_ps(r00,vftabscale);
433 vfitab = _mm256_cvttps_epi32(rt);
434 vfeps = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
435 /* AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
436 vfitab_lo = _mm256_extractf128_si256(vfitab,0x0);
437 vfitab_hi = _mm256_extractf128_si256(vfitab,0x1);
438 vfitab_lo = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
439 vfitab_hi = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
441 /* CUBIC SPLINE TABLE ELECTROSTATICS */
442 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
443 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
444 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
445 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
446 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
447 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
448 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
449 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
450 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
451 Heps = _mm256_mul_ps(vfeps,H);
452 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
453 VV = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
454 velec = _mm256_mul_ps(qq00,VV);
455 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
456 felec = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
458 /* CUBIC SPLINE TABLE DISPERSION */
459 vfitab_lo = _mm_add_epi32(vfitab_lo,ifour);
460 vfitab_hi = _mm_add_epi32(vfitab_hi,ifour);
461 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
462 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
463 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
464 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
465 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
466 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
467 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
468 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
469 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
470 Heps = _mm256_mul_ps(vfeps,H);
471 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
472 VV = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
473 vvdw6 = _mm256_mul_ps(c6_00,VV);
474 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
475 fvdw6 = _mm256_mul_ps(c6_00,FF);
477 /* CUBIC SPLINE TABLE REPULSION */
478 vfitab_lo = _mm_add_epi32(vfitab_lo,ifour);
479 vfitab_hi = _mm_add_epi32(vfitab_hi,ifour);
480 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
481 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
482 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
483 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
484 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
485 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
486 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
487 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
488 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
489 Heps = _mm256_mul_ps(vfeps,H);
490 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
491 VV = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
492 vvdw12 = _mm256_mul_ps(c12_00,VV);
493 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
494 fvdw12 = _mm256_mul_ps(c12_00,FF);
495 vvdw = _mm256_add_ps(vvdw12,vvdw6);
496 fvdw = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
498 /* Update potential sum for this i atom from the interaction with this j atom. */
499 velec = _mm256_andnot_ps(dummy_mask,velec);
500 velecsum = _mm256_add_ps(velecsum,velec);
501 vvdw = _mm256_andnot_ps(dummy_mask,vvdw);
502 vvdwsum = _mm256_add_ps(vvdwsum,vvdw);
504 fscal = _mm256_add_ps(felec,fvdw);
506 fscal = _mm256_andnot_ps(dummy_mask,fscal);
508 /* Calculate temporary vectorial force */
509 tx = _mm256_mul_ps(fscal,dx00);
510 ty = _mm256_mul_ps(fscal,dy00);
511 tz = _mm256_mul_ps(fscal,dz00);
513 /* Update vectorial force */
514 fix0 = _mm256_add_ps(fix0,tx);
515 fiy0 = _mm256_add_ps(fiy0,ty);
516 fiz0 = _mm256_add_ps(fiz0,tz);
518 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
519 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
520 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
521 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
522 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
523 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
524 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
525 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
526 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
528 /* Inner loop uses 74 flops */
531 /* End of innermost loop */
533 gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
534 f+i_coord_offset,fshift+i_shift_offset);
537 /* Update potential energies */
538 gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
539 gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
541 /* Increment number of inner iterations */
542 inneriter += j_index_end - j_index_start;
544 /* Outer loop uses 9 flops */
547 /* Increment number of outer iterations */
550 /* Update outer/inner flops */
552 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*74);
555 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_F_avx_256_single
556 * Electrostatics interaction: CubicSplineTable
557 * VdW interaction: CubicSplineTable
558 * Geometry: Particle-Particle
559 * Calculate force/pot: Force
562 nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_F_avx_256_single
563 (t_nblist * gmx_restrict nlist,
564 rvec * gmx_restrict xx,
565 rvec * gmx_restrict ff,
566 t_forcerec * gmx_restrict fr,
567 t_mdatoms * gmx_restrict mdatoms,
568 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
569 t_nrnb * gmx_restrict nrnb)
571 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
572 * just 0 for non-waters.
573 * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
574 * jnr indices corresponding to data put in the four positions in the SIMD register.
576 int i_shift_offset,i_coord_offset,outeriter,inneriter;
577 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
578 int jnrA,jnrB,jnrC,jnrD;
579 int jnrE,jnrF,jnrG,jnrH;
580 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
581 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
582 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
583 int j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
584 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
586 real *shiftvec,*fshift,*x,*f;
587 real *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
589 __m256 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
590 real * vdwioffsetptr0;
591 __m256 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
592 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
593 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
594 __m256 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
595 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
598 __m256 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
601 __m256 one_sixth = _mm256_set1_ps(1.0/6.0);
602 __m256 one_twelfth = _mm256_set1_ps(1.0/12.0);
604 __m128i vfitab_lo,vfitab_hi;
605 __m128i ifour = _mm_set1_epi32(4);
606 __m256 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
608 __m256 dummy_mask,cutoff_mask;
609 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
610 __m256 one = _mm256_set1_ps(1.0);
611 __m256 two = _mm256_set1_ps(2.0);
617 jindex = nlist->jindex;
619 shiftidx = nlist->shift;
621 shiftvec = fr->shift_vec[0];
622 fshift = fr->fshift[0];
623 facel = _mm256_set1_ps(fr->epsfac);
624 charge = mdatoms->chargeA;
625 nvdwtype = fr->ntype;
627 vdwtype = mdatoms->typeA;
629 vftab = kernel_data->table_elec_vdw->data;
630 vftabscale = _mm256_set1_ps(kernel_data->table_elec_vdw->scale);
632 /* Avoid stupid compiler warnings */
633 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
646 for(iidx=0;iidx<4*DIM;iidx++)
651 /* Start outer loop over neighborlists */
652 for(iidx=0; iidx<nri; iidx++)
654 /* Load shift vector for this list */
655 i_shift_offset = DIM*shiftidx[iidx];
657 /* Load limits for loop over neighbors */
658 j_index_start = jindex[iidx];
659 j_index_end = jindex[iidx+1];
661 /* Get outer coordinate index */
663 i_coord_offset = DIM*inr;
665 /* Load i particle coords and add shift vector */
666 gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
668 fix0 = _mm256_setzero_ps();
669 fiy0 = _mm256_setzero_ps();
670 fiz0 = _mm256_setzero_ps();
672 /* Load parameters for i particles */
673 iq0 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
674 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
676 /* Start inner kernel loop */
677 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
680 /* Get j neighbor index, and coordinate index */
689 j_coord_offsetA = DIM*jnrA;
690 j_coord_offsetB = DIM*jnrB;
691 j_coord_offsetC = DIM*jnrC;
692 j_coord_offsetD = DIM*jnrD;
693 j_coord_offsetE = DIM*jnrE;
694 j_coord_offsetF = DIM*jnrF;
695 j_coord_offsetG = DIM*jnrG;
696 j_coord_offsetH = DIM*jnrH;
698 /* load j atom coordinates */
699 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
700 x+j_coord_offsetC,x+j_coord_offsetD,
701 x+j_coord_offsetE,x+j_coord_offsetF,
702 x+j_coord_offsetG,x+j_coord_offsetH,
705 /* Calculate displacement vector */
706 dx00 = _mm256_sub_ps(ix0,jx0);
707 dy00 = _mm256_sub_ps(iy0,jy0);
708 dz00 = _mm256_sub_ps(iz0,jz0);
710 /* Calculate squared distance and things based on it */
711 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
713 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
715 /* Load parameters for j particles */
716 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
717 charge+jnrC+0,charge+jnrD+0,
718 charge+jnrE+0,charge+jnrF+0,
719 charge+jnrG+0,charge+jnrH+0);
720 vdwjidx0A = 2*vdwtype[jnrA+0];
721 vdwjidx0B = 2*vdwtype[jnrB+0];
722 vdwjidx0C = 2*vdwtype[jnrC+0];
723 vdwjidx0D = 2*vdwtype[jnrD+0];
724 vdwjidx0E = 2*vdwtype[jnrE+0];
725 vdwjidx0F = 2*vdwtype[jnrF+0];
726 vdwjidx0G = 2*vdwtype[jnrG+0];
727 vdwjidx0H = 2*vdwtype[jnrH+0];
729 /**************************
730 * CALCULATE INTERACTIONS *
731 **************************/
733 r00 = _mm256_mul_ps(rsq00,rinv00);
735 /* Compute parameters for interactions between i and j atoms */
736 qq00 = _mm256_mul_ps(iq0,jq0);
737 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
738 vdwioffsetptr0+vdwjidx0B,
739 vdwioffsetptr0+vdwjidx0C,
740 vdwioffsetptr0+vdwjidx0D,
741 vdwioffsetptr0+vdwjidx0E,
742 vdwioffsetptr0+vdwjidx0F,
743 vdwioffsetptr0+vdwjidx0G,
744 vdwioffsetptr0+vdwjidx0H,
747 /* Calculate table index by multiplying r with table scale and truncate to integer */
748 rt = _mm256_mul_ps(r00,vftabscale);
749 vfitab = _mm256_cvttps_epi32(rt);
750 vfeps = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
751 /* AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
752 vfitab_lo = _mm256_extractf128_si256(vfitab,0x0);
753 vfitab_hi = _mm256_extractf128_si256(vfitab,0x1);
754 vfitab_lo = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
755 vfitab_hi = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
757 /* CUBIC SPLINE TABLE ELECTROSTATICS */
758 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
759 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
760 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
761 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
762 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
763 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
764 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
765 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
766 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
767 Heps = _mm256_mul_ps(vfeps,H);
768 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
769 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
770 felec = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
772 /* CUBIC SPLINE TABLE DISPERSION */
773 vfitab_lo = _mm_add_epi32(vfitab_lo,ifour);
774 vfitab_hi = _mm_add_epi32(vfitab_hi,ifour);
775 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
776 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
777 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
778 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
779 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
780 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
781 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
782 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
783 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
784 Heps = _mm256_mul_ps(vfeps,H);
785 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
786 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
787 fvdw6 = _mm256_mul_ps(c6_00,FF);
789 /* CUBIC SPLINE TABLE REPULSION */
790 vfitab_lo = _mm_add_epi32(vfitab_lo,ifour);
791 vfitab_hi = _mm_add_epi32(vfitab_hi,ifour);
792 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
793 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
794 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
795 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
796 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
797 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
798 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
799 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
800 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
801 Heps = _mm256_mul_ps(vfeps,H);
802 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
803 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
804 fvdw12 = _mm256_mul_ps(c12_00,FF);
805 fvdw = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
807 fscal = _mm256_add_ps(felec,fvdw);
809 /* Calculate temporary vectorial force */
810 tx = _mm256_mul_ps(fscal,dx00);
811 ty = _mm256_mul_ps(fscal,dy00);
812 tz = _mm256_mul_ps(fscal,dz00);
814 /* Update vectorial force */
815 fix0 = _mm256_add_ps(fix0,tx);
816 fiy0 = _mm256_add_ps(fiy0,ty);
817 fiz0 = _mm256_add_ps(fiz0,tz);
819 fjptrA = f+j_coord_offsetA;
820 fjptrB = f+j_coord_offsetB;
821 fjptrC = f+j_coord_offsetC;
822 fjptrD = f+j_coord_offsetD;
823 fjptrE = f+j_coord_offsetE;
824 fjptrF = f+j_coord_offsetF;
825 fjptrG = f+j_coord_offsetG;
826 fjptrH = f+j_coord_offsetH;
827 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
829 /* Inner loop uses 61 flops */
835 /* Get j neighbor index, and coordinate index */
836 jnrlistA = jjnr[jidx];
837 jnrlistB = jjnr[jidx+1];
838 jnrlistC = jjnr[jidx+2];
839 jnrlistD = jjnr[jidx+3];
840 jnrlistE = jjnr[jidx+4];
841 jnrlistF = jjnr[jidx+5];
842 jnrlistG = jjnr[jidx+6];
843 jnrlistH = jjnr[jidx+7];
844 /* Sign of each element will be negative for non-real atoms.
845 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
846 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
848 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
849 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
851 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
852 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
853 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
854 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
855 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
856 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
857 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
858 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
859 j_coord_offsetA = DIM*jnrA;
860 j_coord_offsetB = DIM*jnrB;
861 j_coord_offsetC = DIM*jnrC;
862 j_coord_offsetD = DIM*jnrD;
863 j_coord_offsetE = DIM*jnrE;
864 j_coord_offsetF = DIM*jnrF;
865 j_coord_offsetG = DIM*jnrG;
866 j_coord_offsetH = DIM*jnrH;
868 /* load j atom coordinates */
869 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
870 x+j_coord_offsetC,x+j_coord_offsetD,
871 x+j_coord_offsetE,x+j_coord_offsetF,
872 x+j_coord_offsetG,x+j_coord_offsetH,
875 /* Calculate displacement vector */
876 dx00 = _mm256_sub_ps(ix0,jx0);
877 dy00 = _mm256_sub_ps(iy0,jy0);
878 dz00 = _mm256_sub_ps(iz0,jz0);
880 /* Calculate squared distance and things based on it */
881 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
883 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
885 /* Load parameters for j particles */
886 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
887 charge+jnrC+0,charge+jnrD+0,
888 charge+jnrE+0,charge+jnrF+0,
889 charge+jnrG+0,charge+jnrH+0);
890 vdwjidx0A = 2*vdwtype[jnrA+0];
891 vdwjidx0B = 2*vdwtype[jnrB+0];
892 vdwjidx0C = 2*vdwtype[jnrC+0];
893 vdwjidx0D = 2*vdwtype[jnrD+0];
894 vdwjidx0E = 2*vdwtype[jnrE+0];
895 vdwjidx0F = 2*vdwtype[jnrF+0];
896 vdwjidx0G = 2*vdwtype[jnrG+0];
897 vdwjidx0H = 2*vdwtype[jnrH+0];
899 /**************************
900 * CALCULATE INTERACTIONS *
901 **************************/
903 r00 = _mm256_mul_ps(rsq00,rinv00);
904 r00 = _mm256_andnot_ps(dummy_mask,r00);
906 /* Compute parameters for interactions between i and j atoms */
907 qq00 = _mm256_mul_ps(iq0,jq0);
908 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
909 vdwioffsetptr0+vdwjidx0B,
910 vdwioffsetptr0+vdwjidx0C,
911 vdwioffsetptr0+vdwjidx0D,
912 vdwioffsetptr0+vdwjidx0E,
913 vdwioffsetptr0+vdwjidx0F,
914 vdwioffsetptr0+vdwjidx0G,
915 vdwioffsetptr0+vdwjidx0H,
918 /* Calculate table index by multiplying r with table scale and truncate to integer */
919 rt = _mm256_mul_ps(r00,vftabscale);
920 vfitab = _mm256_cvttps_epi32(rt);
921 vfeps = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
922 /* AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
923 vfitab_lo = _mm256_extractf128_si256(vfitab,0x0);
924 vfitab_hi = _mm256_extractf128_si256(vfitab,0x1);
925 vfitab_lo = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
926 vfitab_hi = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
928 /* CUBIC SPLINE TABLE ELECTROSTATICS */
929 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
930 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
931 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
932 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
933 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
934 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
935 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
936 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
937 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
938 Heps = _mm256_mul_ps(vfeps,H);
939 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
940 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
941 felec = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
943 /* CUBIC SPLINE TABLE DISPERSION */
944 vfitab_lo = _mm_add_epi32(vfitab_lo,ifour);
945 vfitab_hi = _mm_add_epi32(vfitab_hi,ifour);
946 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
947 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
948 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
949 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
950 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
951 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
952 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
953 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
954 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
955 Heps = _mm256_mul_ps(vfeps,H);
956 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
957 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
958 fvdw6 = _mm256_mul_ps(c6_00,FF);
960 /* CUBIC SPLINE TABLE REPULSION */
961 vfitab_lo = _mm_add_epi32(vfitab_lo,ifour);
962 vfitab_hi = _mm_add_epi32(vfitab_hi,ifour);
963 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
964 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
965 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
966 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
967 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
968 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
969 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
970 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
971 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
972 Heps = _mm256_mul_ps(vfeps,H);
973 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
974 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
975 fvdw12 = _mm256_mul_ps(c12_00,FF);
976 fvdw = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
978 fscal = _mm256_add_ps(felec,fvdw);
980 fscal = _mm256_andnot_ps(dummy_mask,fscal);
982 /* Calculate temporary vectorial force */
983 tx = _mm256_mul_ps(fscal,dx00);
984 ty = _mm256_mul_ps(fscal,dy00);
985 tz = _mm256_mul_ps(fscal,dz00);
987 /* Update vectorial force */
988 fix0 = _mm256_add_ps(fix0,tx);
989 fiy0 = _mm256_add_ps(fiy0,ty);
990 fiz0 = _mm256_add_ps(fiz0,tz);
992 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
993 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
994 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
995 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
996 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
997 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
998 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
999 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1000 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
1002 /* Inner loop uses 62 flops */
1005 /* End of innermost loop */
1007 gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
1008 f+i_coord_offset,fshift+i_shift_offset);
1010 /* Increment number of inner iterations */
1011 inneriter += j_index_end - j_index_start;
1013 /* Outer loop uses 7 flops */
1016 /* Increment number of outer iterations */
1019 /* Update outer/inner flops */
1021 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*62);