2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
5 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
6 * and including many others, as listed in the AUTHORS file in the
7 * top-level source directory and at http://www.gromacs.org.
9 * GROMACS is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public License
11 * as published by the Free Software Foundation; either version 2.1
12 * of the License, or (at your option) any later version.
14 * GROMACS is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with GROMACS; if not, see
21 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 * If you want to redistribute modifications to GROMACS, please
25 * consider that scientific software is very special. Version
26 * control is crucial - bugs must be traceable. We will be happy to
27 * consider code for inclusion in the official distribution, but
28 * derived work must not be called official GROMACS. Details are found
29 * in the README & COPYING files - if they are missing, get the
30 * official version at http://www.gromacs.org.
32 * To help us fund GROMACS development, we humbly ask that you cite
33 * the research papers on the package. Check out http://www.gromacs.org.
36 * Note: this file was generated by the GROMACS avx_128_fma_single kernel generator.
42 #include "../nb_kernel.h"
43 #include "types/simple.h"
44 #include "gromacs/math/vec.h"
47 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
48 #include "kernelutil_x86_avx_128_fma_single.h"
51 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_VF_avx_128_fma_single
52 * Electrostatics interaction: CubicSplineTable
53 * VdW interaction: LennardJones
54 * Geometry: Particle-Particle
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_VF_avx_128_fma_single
59 (t_nblist * gmx_restrict nlist,
60 rvec * gmx_restrict xx,
61 rvec * gmx_restrict ff,
62 t_forcerec * gmx_restrict fr,
63 t_mdatoms * gmx_restrict mdatoms,
64 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
65 t_nrnb * gmx_restrict nrnb)
67 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
68 * just 0 for non-waters.
69 * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
70 * jnr indices corresponding to data put in the four positions in the SIMD register.
72 int i_shift_offset,i_coord_offset,outeriter,inneriter;
73 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
74 int jnrA,jnrB,jnrC,jnrD;
75 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
76 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
77 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
79 real *shiftvec,*fshift,*x,*f;
80 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
82 __m128 fscal,rcutoff,rcutoff2,jidxall;
84 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
85 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
86 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
87 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
88 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
91 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
94 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
95 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
97 __m128i ifour = _mm_set1_epi32(4);
98 __m128 rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
100 __m128 dummy_mask,cutoff_mask;
101 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
102 __m128 one = _mm_set1_ps(1.0);
103 __m128 two = _mm_set1_ps(2.0);
109 jindex = nlist->jindex;
111 shiftidx = nlist->shift;
113 shiftvec = fr->shift_vec[0];
114 fshift = fr->fshift[0];
115 facel = _mm_set1_ps(fr->epsfac);
116 charge = mdatoms->chargeA;
117 nvdwtype = fr->ntype;
119 vdwtype = mdatoms->typeA;
121 vftab = kernel_data->table_elec->data;
122 vftabscale = _mm_set1_ps(kernel_data->table_elec->scale);
124 /* Avoid stupid compiler warnings */
125 jnrA = jnrB = jnrC = jnrD = 0;
134 for(iidx=0;iidx<4*DIM;iidx++)
139 /* Start outer loop over neighborlists */
140 for(iidx=0; iidx<nri; iidx++)
142 /* Load shift vector for this list */
143 i_shift_offset = DIM*shiftidx[iidx];
145 /* Load limits for loop over neighbors */
146 j_index_start = jindex[iidx];
147 j_index_end = jindex[iidx+1];
149 /* Get outer coordinate index */
151 i_coord_offset = DIM*inr;
153 /* Load i particle coords and add shift vector */
154 gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
156 fix0 = _mm_setzero_ps();
157 fiy0 = _mm_setzero_ps();
158 fiz0 = _mm_setzero_ps();
160 /* Load parameters for i particles */
161 iq0 = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
162 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
164 /* Reset potential sums */
165 velecsum = _mm_setzero_ps();
166 vvdwsum = _mm_setzero_ps();
168 /* Start inner kernel loop */
169 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
172 /* Get j neighbor index, and coordinate index */
177 j_coord_offsetA = DIM*jnrA;
178 j_coord_offsetB = DIM*jnrB;
179 j_coord_offsetC = DIM*jnrC;
180 j_coord_offsetD = DIM*jnrD;
182 /* load j atom coordinates */
183 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
184 x+j_coord_offsetC,x+j_coord_offsetD,
187 /* Calculate displacement vector */
188 dx00 = _mm_sub_ps(ix0,jx0);
189 dy00 = _mm_sub_ps(iy0,jy0);
190 dz00 = _mm_sub_ps(iz0,jz0);
192 /* Calculate squared distance and things based on it */
193 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
195 rinv00 = gmx_mm_invsqrt_ps(rsq00);
197 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
199 /* Load parameters for j particles */
200 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
201 charge+jnrC+0,charge+jnrD+0);
202 vdwjidx0A = 2*vdwtype[jnrA+0];
203 vdwjidx0B = 2*vdwtype[jnrB+0];
204 vdwjidx0C = 2*vdwtype[jnrC+0];
205 vdwjidx0D = 2*vdwtype[jnrD+0];
207 /**************************
208 * CALCULATE INTERACTIONS *
209 **************************/
211 r00 = _mm_mul_ps(rsq00,rinv00);
213 /* Compute parameters for interactions between i and j atoms */
214 qq00 = _mm_mul_ps(iq0,jq0);
215 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
216 vdwparam+vdwioffset0+vdwjidx0B,
217 vdwparam+vdwioffset0+vdwjidx0C,
218 vdwparam+vdwioffset0+vdwjidx0D,
221 /* Calculate table index by multiplying r with table scale and truncate to integer */
222 rt = _mm_mul_ps(r00,vftabscale);
223 vfitab = _mm_cvttps_epi32(rt);
225 vfeps = _mm_frcz_ps(rt);
227 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
229 twovfeps = _mm_add_ps(vfeps,vfeps);
230 vfitab = _mm_slli_epi32(vfitab,2);
232 /* CUBIC SPLINE TABLE ELECTROSTATICS */
233 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
234 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
235 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
236 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
237 _MM_TRANSPOSE4_PS(Y,F,G,H);
238 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
239 VV = _mm_macc_ps(vfeps,Fp,Y);
240 velec = _mm_mul_ps(qq00,VV);
241 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
242 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
244 /* LENNARD-JONES DISPERSION/REPULSION */
246 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
247 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
248 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
249 vvdw = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
250 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
252 /* Update potential sum for this i atom from the interaction with this j atom. */
253 velecsum = _mm_add_ps(velecsum,velec);
254 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
256 fscal = _mm_add_ps(felec,fvdw);
258 /* Update vectorial force */
259 fix0 = _mm_macc_ps(dx00,fscal,fix0);
260 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
261 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
263 fjptrA = f+j_coord_offsetA;
264 fjptrB = f+j_coord_offsetB;
265 fjptrC = f+j_coord_offsetC;
266 fjptrD = f+j_coord_offsetD;
267 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
268 _mm_mul_ps(dx00,fscal),
269 _mm_mul_ps(dy00,fscal),
270 _mm_mul_ps(dz00,fscal));
272 /* Inner loop uses 59 flops */
278 /* Get j neighbor index, and coordinate index */
279 jnrlistA = jjnr[jidx];
280 jnrlistB = jjnr[jidx+1];
281 jnrlistC = jjnr[jidx+2];
282 jnrlistD = jjnr[jidx+3];
283 /* Sign of each element will be negative for non-real atoms.
284 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
285 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
287 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
288 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
289 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
290 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
291 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
292 j_coord_offsetA = DIM*jnrA;
293 j_coord_offsetB = DIM*jnrB;
294 j_coord_offsetC = DIM*jnrC;
295 j_coord_offsetD = DIM*jnrD;
297 /* load j atom coordinates */
298 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
299 x+j_coord_offsetC,x+j_coord_offsetD,
302 /* Calculate displacement vector */
303 dx00 = _mm_sub_ps(ix0,jx0);
304 dy00 = _mm_sub_ps(iy0,jy0);
305 dz00 = _mm_sub_ps(iz0,jz0);
307 /* Calculate squared distance and things based on it */
308 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
310 rinv00 = gmx_mm_invsqrt_ps(rsq00);
312 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
314 /* Load parameters for j particles */
315 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
316 charge+jnrC+0,charge+jnrD+0);
317 vdwjidx0A = 2*vdwtype[jnrA+0];
318 vdwjidx0B = 2*vdwtype[jnrB+0];
319 vdwjidx0C = 2*vdwtype[jnrC+0];
320 vdwjidx0D = 2*vdwtype[jnrD+0];
322 /**************************
323 * CALCULATE INTERACTIONS *
324 **************************/
326 r00 = _mm_mul_ps(rsq00,rinv00);
327 r00 = _mm_andnot_ps(dummy_mask,r00);
329 /* Compute parameters for interactions between i and j atoms */
330 qq00 = _mm_mul_ps(iq0,jq0);
331 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
332 vdwparam+vdwioffset0+vdwjidx0B,
333 vdwparam+vdwioffset0+vdwjidx0C,
334 vdwparam+vdwioffset0+vdwjidx0D,
337 /* Calculate table index by multiplying r with table scale and truncate to integer */
338 rt = _mm_mul_ps(r00,vftabscale);
339 vfitab = _mm_cvttps_epi32(rt);
341 vfeps = _mm_frcz_ps(rt);
343 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
345 twovfeps = _mm_add_ps(vfeps,vfeps);
346 vfitab = _mm_slli_epi32(vfitab,2);
348 /* CUBIC SPLINE TABLE ELECTROSTATICS */
349 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
350 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
351 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
352 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
353 _MM_TRANSPOSE4_PS(Y,F,G,H);
354 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
355 VV = _mm_macc_ps(vfeps,Fp,Y);
356 velec = _mm_mul_ps(qq00,VV);
357 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
358 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
360 /* LENNARD-JONES DISPERSION/REPULSION */
362 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
363 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
364 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
365 vvdw = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
366 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
368 /* Update potential sum for this i atom from the interaction with this j atom. */
369 velec = _mm_andnot_ps(dummy_mask,velec);
370 velecsum = _mm_add_ps(velecsum,velec);
371 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
372 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
374 fscal = _mm_add_ps(felec,fvdw);
376 fscal = _mm_andnot_ps(dummy_mask,fscal);
378 /* Update vectorial force */
379 fix0 = _mm_macc_ps(dx00,fscal,fix0);
380 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
381 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
383 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
384 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
385 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
386 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
387 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
388 _mm_mul_ps(dx00,fscal),
389 _mm_mul_ps(dy00,fscal),
390 _mm_mul_ps(dz00,fscal));
392 /* Inner loop uses 60 flops */
395 /* End of innermost loop */
397 gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
398 f+i_coord_offset,fshift+i_shift_offset);
401 /* Update potential energies */
402 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
403 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
405 /* Increment number of inner iterations */
406 inneriter += j_index_end - j_index_start;
408 /* Outer loop uses 9 flops */
411 /* Increment number of outer iterations */
414 /* Update outer/inner flops */
416 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*60);
419 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_F_avx_128_fma_single
420 * Electrostatics interaction: CubicSplineTable
421 * VdW interaction: LennardJones
422 * Geometry: Particle-Particle
423 * Calculate force/pot: Force
426 nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_F_avx_128_fma_single
427 (t_nblist * gmx_restrict nlist,
428 rvec * gmx_restrict xx,
429 rvec * gmx_restrict ff,
430 t_forcerec * gmx_restrict fr,
431 t_mdatoms * gmx_restrict mdatoms,
432 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
433 t_nrnb * gmx_restrict nrnb)
435 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
436 * just 0 for non-waters.
437 * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
438 * jnr indices corresponding to data put in the four positions in the SIMD register.
440 int i_shift_offset,i_coord_offset,outeriter,inneriter;
441 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
442 int jnrA,jnrB,jnrC,jnrD;
443 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
444 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
445 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
447 real *shiftvec,*fshift,*x,*f;
448 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
450 __m128 fscal,rcutoff,rcutoff2,jidxall;
452 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
453 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
454 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
455 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
456 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
459 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
462 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
463 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
465 __m128i ifour = _mm_set1_epi32(4);
466 __m128 rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
468 __m128 dummy_mask,cutoff_mask;
469 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
470 __m128 one = _mm_set1_ps(1.0);
471 __m128 two = _mm_set1_ps(2.0);
477 jindex = nlist->jindex;
479 shiftidx = nlist->shift;
481 shiftvec = fr->shift_vec[0];
482 fshift = fr->fshift[0];
483 facel = _mm_set1_ps(fr->epsfac);
484 charge = mdatoms->chargeA;
485 nvdwtype = fr->ntype;
487 vdwtype = mdatoms->typeA;
489 vftab = kernel_data->table_elec->data;
490 vftabscale = _mm_set1_ps(kernel_data->table_elec->scale);
492 /* Avoid stupid compiler warnings */
493 jnrA = jnrB = jnrC = jnrD = 0;
502 for(iidx=0;iidx<4*DIM;iidx++)
507 /* Start outer loop over neighborlists */
508 for(iidx=0; iidx<nri; iidx++)
510 /* Load shift vector for this list */
511 i_shift_offset = DIM*shiftidx[iidx];
513 /* Load limits for loop over neighbors */
514 j_index_start = jindex[iidx];
515 j_index_end = jindex[iidx+1];
517 /* Get outer coordinate index */
519 i_coord_offset = DIM*inr;
521 /* Load i particle coords and add shift vector */
522 gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
524 fix0 = _mm_setzero_ps();
525 fiy0 = _mm_setzero_ps();
526 fiz0 = _mm_setzero_ps();
528 /* Load parameters for i particles */
529 iq0 = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
530 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
532 /* Start inner kernel loop */
533 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
536 /* Get j neighbor index, and coordinate index */
541 j_coord_offsetA = DIM*jnrA;
542 j_coord_offsetB = DIM*jnrB;
543 j_coord_offsetC = DIM*jnrC;
544 j_coord_offsetD = DIM*jnrD;
546 /* load j atom coordinates */
547 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
548 x+j_coord_offsetC,x+j_coord_offsetD,
551 /* Calculate displacement vector */
552 dx00 = _mm_sub_ps(ix0,jx0);
553 dy00 = _mm_sub_ps(iy0,jy0);
554 dz00 = _mm_sub_ps(iz0,jz0);
556 /* Calculate squared distance and things based on it */
557 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
559 rinv00 = gmx_mm_invsqrt_ps(rsq00);
561 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
563 /* Load parameters for j particles */
564 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
565 charge+jnrC+0,charge+jnrD+0);
566 vdwjidx0A = 2*vdwtype[jnrA+0];
567 vdwjidx0B = 2*vdwtype[jnrB+0];
568 vdwjidx0C = 2*vdwtype[jnrC+0];
569 vdwjidx0D = 2*vdwtype[jnrD+0];
571 /**************************
572 * CALCULATE INTERACTIONS *
573 **************************/
575 r00 = _mm_mul_ps(rsq00,rinv00);
577 /* Compute parameters for interactions between i and j atoms */
578 qq00 = _mm_mul_ps(iq0,jq0);
579 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
580 vdwparam+vdwioffset0+vdwjidx0B,
581 vdwparam+vdwioffset0+vdwjidx0C,
582 vdwparam+vdwioffset0+vdwjidx0D,
585 /* Calculate table index by multiplying r with table scale and truncate to integer */
586 rt = _mm_mul_ps(r00,vftabscale);
587 vfitab = _mm_cvttps_epi32(rt);
589 vfeps = _mm_frcz_ps(rt);
591 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
593 twovfeps = _mm_add_ps(vfeps,vfeps);
594 vfitab = _mm_slli_epi32(vfitab,2);
596 /* CUBIC SPLINE TABLE ELECTROSTATICS */
597 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
598 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
599 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
600 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
601 _MM_TRANSPOSE4_PS(Y,F,G,H);
602 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
603 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
604 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
606 /* LENNARD-JONES DISPERSION/REPULSION */
608 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
609 fvdw = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
611 fscal = _mm_add_ps(felec,fvdw);
613 /* Update vectorial force */
614 fix0 = _mm_macc_ps(dx00,fscal,fix0);
615 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
616 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
618 fjptrA = f+j_coord_offsetA;
619 fjptrB = f+j_coord_offsetB;
620 fjptrC = f+j_coord_offsetC;
621 fjptrD = f+j_coord_offsetD;
622 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
623 _mm_mul_ps(dx00,fscal),
624 _mm_mul_ps(dy00,fscal),
625 _mm_mul_ps(dz00,fscal));
627 /* Inner loop uses 50 flops */
633 /* Get j neighbor index, and coordinate index */
634 jnrlistA = jjnr[jidx];
635 jnrlistB = jjnr[jidx+1];
636 jnrlistC = jjnr[jidx+2];
637 jnrlistD = jjnr[jidx+3];
638 /* Sign of each element will be negative for non-real atoms.
639 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
640 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
642 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
643 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
644 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
645 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
646 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
647 j_coord_offsetA = DIM*jnrA;
648 j_coord_offsetB = DIM*jnrB;
649 j_coord_offsetC = DIM*jnrC;
650 j_coord_offsetD = DIM*jnrD;
652 /* load j atom coordinates */
653 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
654 x+j_coord_offsetC,x+j_coord_offsetD,
657 /* Calculate displacement vector */
658 dx00 = _mm_sub_ps(ix0,jx0);
659 dy00 = _mm_sub_ps(iy0,jy0);
660 dz00 = _mm_sub_ps(iz0,jz0);
662 /* Calculate squared distance and things based on it */
663 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
665 rinv00 = gmx_mm_invsqrt_ps(rsq00);
667 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
669 /* Load parameters for j particles */
670 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
671 charge+jnrC+0,charge+jnrD+0);
672 vdwjidx0A = 2*vdwtype[jnrA+0];
673 vdwjidx0B = 2*vdwtype[jnrB+0];
674 vdwjidx0C = 2*vdwtype[jnrC+0];
675 vdwjidx0D = 2*vdwtype[jnrD+0];
677 /**************************
678 * CALCULATE INTERACTIONS *
679 **************************/
681 r00 = _mm_mul_ps(rsq00,rinv00);
682 r00 = _mm_andnot_ps(dummy_mask,r00);
684 /* Compute parameters for interactions between i and j atoms */
685 qq00 = _mm_mul_ps(iq0,jq0);
686 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
687 vdwparam+vdwioffset0+vdwjidx0B,
688 vdwparam+vdwioffset0+vdwjidx0C,
689 vdwparam+vdwioffset0+vdwjidx0D,
692 /* Calculate table index by multiplying r with table scale and truncate to integer */
693 rt = _mm_mul_ps(r00,vftabscale);
694 vfitab = _mm_cvttps_epi32(rt);
696 vfeps = _mm_frcz_ps(rt);
698 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
700 twovfeps = _mm_add_ps(vfeps,vfeps);
701 vfitab = _mm_slli_epi32(vfitab,2);
703 /* CUBIC SPLINE TABLE ELECTROSTATICS */
704 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
705 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
706 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
707 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
708 _MM_TRANSPOSE4_PS(Y,F,G,H);
709 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
710 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
711 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
713 /* LENNARD-JONES DISPERSION/REPULSION */
715 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
716 fvdw = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
718 fscal = _mm_add_ps(felec,fvdw);
720 fscal = _mm_andnot_ps(dummy_mask,fscal);
722 /* Update vectorial force */
723 fix0 = _mm_macc_ps(dx00,fscal,fix0);
724 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
725 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
727 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
728 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
729 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
730 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
731 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
732 _mm_mul_ps(dx00,fscal),
733 _mm_mul_ps(dy00,fscal),
734 _mm_mul_ps(dz00,fscal));
736 /* Inner loop uses 51 flops */
739 /* End of innermost loop */
741 gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
742 f+i_coord_offset,fshift+i_shift_offset);
744 /* Increment number of inner iterations */
745 inneriter += j_index_end - j_index_start;
747 /* Outer loop uses 7 flops */
750 /* Increment number of outer iterations */
753 /* Update outer/inner flops */
755 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*51);