2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014,2015,2017,2018, 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.
44 #include "../nb_kernel.h"
45 #include "gromacs/gmxlib/nrnb.h"
47 #include "kernelutil_x86_avx_128_fma_single.h"
50 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwNone_GeomW4W4_VF_avx_128_fma_single
51 * Electrostatics interaction: CubicSplineTable
52 * VdW interaction: None
53 * Geometry: Water4-Water4
54 * Calculate force/pot: PotentialAndForce
57 nb_kernel_ElecCSTab_VdwNone_GeomW4W4_VF_avx_128_fma_single
58 (t_nblist * gmx_restrict nlist,
59 rvec * gmx_restrict xx,
60 rvec * gmx_restrict ff,
61 struct t_forcerec * gmx_restrict fr,
62 t_mdatoms * gmx_restrict mdatoms,
63 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
64 t_nrnb * gmx_restrict nrnb)
66 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
67 * just 0 for non-waters.
68 * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
69 * jnr indices corresponding to data put in the four positions in the SIMD register.
71 int i_shift_offset,i_coord_offset,outeriter,inneriter;
72 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
73 int jnrA,jnrB,jnrC,jnrD;
74 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
75 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
76 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
78 real *shiftvec,*fshift,*x,*f;
79 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
81 __m128 fscal,rcutoff,rcutoff2,jidxall;
83 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
85 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
87 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
88 int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
89 __m128 jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
90 int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
91 __m128 jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
92 int vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
93 __m128 jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
94 __m128 dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
95 __m128 dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
96 __m128 dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
97 __m128 dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
98 __m128 dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
99 __m128 dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
100 __m128 dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
101 __m128 dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
102 __m128 dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
103 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
106 __m128i ifour = _mm_set1_epi32(4);
107 __m128 rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
109 __m128 dummy_mask,cutoff_mask;
110 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
111 __m128 one = _mm_set1_ps(1.0);
112 __m128 two = _mm_set1_ps(2.0);
118 jindex = nlist->jindex;
120 shiftidx = nlist->shift;
122 shiftvec = fr->shift_vec[0];
123 fshift = fr->fshift[0];
124 facel = _mm_set1_ps(fr->ic->epsfac);
125 charge = mdatoms->chargeA;
127 vftab = kernel_data->table_elec->data;
128 vftabscale = _mm_set1_ps(kernel_data->table_elec->scale);
130 /* Setup water-specific parameters */
131 inr = nlist->iinr[0];
132 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
133 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
134 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
136 jq1 = _mm_set1_ps(charge[inr+1]);
137 jq2 = _mm_set1_ps(charge[inr+2]);
138 jq3 = _mm_set1_ps(charge[inr+3]);
139 qq11 = _mm_mul_ps(iq1,jq1);
140 qq12 = _mm_mul_ps(iq1,jq2);
141 qq13 = _mm_mul_ps(iq1,jq3);
142 qq21 = _mm_mul_ps(iq2,jq1);
143 qq22 = _mm_mul_ps(iq2,jq2);
144 qq23 = _mm_mul_ps(iq2,jq3);
145 qq31 = _mm_mul_ps(iq3,jq1);
146 qq32 = _mm_mul_ps(iq3,jq2);
147 qq33 = _mm_mul_ps(iq3,jq3);
149 /* Avoid stupid compiler warnings */
150 jnrA = jnrB = jnrC = jnrD = 0;
159 for(iidx=0;iidx<4*DIM;iidx++)
164 /* Start outer loop over neighborlists */
165 for(iidx=0; iidx<nri; iidx++)
167 /* Load shift vector for this list */
168 i_shift_offset = DIM*shiftidx[iidx];
170 /* Load limits for loop over neighbors */
171 j_index_start = jindex[iidx];
172 j_index_end = jindex[iidx+1];
174 /* Get outer coordinate index */
176 i_coord_offset = DIM*inr;
178 /* Load i particle coords and add shift vector */
179 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
180 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
182 fix1 = _mm_setzero_ps();
183 fiy1 = _mm_setzero_ps();
184 fiz1 = _mm_setzero_ps();
185 fix2 = _mm_setzero_ps();
186 fiy2 = _mm_setzero_ps();
187 fiz2 = _mm_setzero_ps();
188 fix3 = _mm_setzero_ps();
189 fiy3 = _mm_setzero_ps();
190 fiz3 = _mm_setzero_ps();
192 /* Reset potential sums */
193 velecsum = _mm_setzero_ps();
195 /* Start inner kernel loop */
196 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
199 /* Get j neighbor index, and coordinate index */
204 j_coord_offsetA = DIM*jnrA;
205 j_coord_offsetB = DIM*jnrB;
206 j_coord_offsetC = DIM*jnrC;
207 j_coord_offsetD = DIM*jnrD;
209 /* load j atom coordinates */
210 gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
211 x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
212 &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
214 /* Calculate displacement vector */
215 dx11 = _mm_sub_ps(ix1,jx1);
216 dy11 = _mm_sub_ps(iy1,jy1);
217 dz11 = _mm_sub_ps(iz1,jz1);
218 dx12 = _mm_sub_ps(ix1,jx2);
219 dy12 = _mm_sub_ps(iy1,jy2);
220 dz12 = _mm_sub_ps(iz1,jz2);
221 dx13 = _mm_sub_ps(ix1,jx3);
222 dy13 = _mm_sub_ps(iy1,jy3);
223 dz13 = _mm_sub_ps(iz1,jz3);
224 dx21 = _mm_sub_ps(ix2,jx1);
225 dy21 = _mm_sub_ps(iy2,jy1);
226 dz21 = _mm_sub_ps(iz2,jz1);
227 dx22 = _mm_sub_ps(ix2,jx2);
228 dy22 = _mm_sub_ps(iy2,jy2);
229 dz22 = _mm_sub_ps(iz2,jz2);
230 dx23 = _mm_sub_ps(ix2,jx3);
231 dy23 = _mm_sub_ps(iy2,jy3);
232 dz23 = _mm_sub_ps(iz2,jz3);
233 dx31 = _mm_sub_ps(ix3,jx1);
234 dy31 = _mm_sub_ps(iy3,jy1);
235 dz31 = _mm_sub_ps(iz3,jz1);
236 dx32 = _mm_sub_ps(ix3,jx2);
237 dy32 = _mm_sub_ps(iy3,jy2);
238 dz32 = _mm_sub_ps(iz3,jz2);
239 dx33 = _mm_sub_ps(ix3,jx3);
240 dy33 = _mm_sub_ps(iy3,jy3);
241 dz33 = _mm_sub_ps(iz3,jz3);
243 /* Calculate squared distance and things based on it */
244 rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
245 rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
246 rsq13 = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
247 rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
248 rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
249 rsq23 = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
250 rsq31 = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
251 rsq32 = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
252 rsq33 = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
254 rinv11 = avx128fma_invsqrt_f(rsq11);
255 rinv12 = avx128fma_invsqrt_f(rsq12);
256 rinv13 = avx128fma_invsqrt_f(rsq13);
257 rinv21 = avx128fma_invsqrt_f(rsq21);
258 rinv22 = avx128fma_invsqrt_f(rsq22);
259 rinv23 = avx128fma_invsqrt_f(rsq23);
260 rinv31 = avx128fma_invsqrt_f(rsq31);
261 rinv32 = avx128fma_invsqrt_f(rsq32);
262 rinv33 = avx128fma_invsqrt_f(rsq33);
264 fjx1 = _mm_setzero_ps();
265 fjy1 = _mm_setzero_ps();
266 fjz1 = _mm_setzero_ps();
267 fjx2 = _mm_setzero_ps();
268 fjy2 = _mm_setzero_ps();
269 fjz2 = _mm_setzero_ps();
270 fjx3 = _mm_setzero_ps();
271 fjy3 = _mm_setzero_ps();
272 fjz3 = _mm_setzero_ps();
274 /**************************
275 * CALCULATE INTERACTIONS *
276 **************************/
278 r11 = _mm_mul_ps(rsq11,rinv11);
280 /* Calculate table index by multiplying r with table scale and truncate to integer */
281 rt = _mm_mul_ps(r11,vftabscale);
282 vfitab = _mm_cvttps_epi32(rt);
284 vfeps = _mm_frcz_ps(rt);
286 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
288 twovfeps = _mm_add_ps(vfeps,vfeps);
289 vfitab = _mm_slli_epi32(vfitab,2);
291 /* CUBIC SPLINE TABLE ELECTROSTATICS */
292 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
293 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
294 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
295 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
296 _MM_TRANSPOSE4_PS(Y,F,G,H);
297 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
298 VV = _mm_macc_ps(vfeps,Fp,Y);
299 velec = _mm_mul_ps(qq11,VV);
300 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
301 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq11,FF),_mm_mul_ps(vftabscale,rinv11)));
303 /* Update potential sum for this i atom from the interaction with this j atom. */
304 velecsum = _mm_add_ps(velecsum,velec);
308 /* Update vectorial force */
309 fix1 = _mm_macc_ps(dx11,fscal,fix1);
310 fiy1 = _mm_macc_ps(dy11,fscal,fiy1);
311 fiz1 = _mm_macc_ps(dz11,fscal,fiz1);
313 fjx1 = _mm_macc_ps(dx11,fscal,fjx1);
314 fjy1 = _mm_macc_ps(dy11,fscal,fjy1);
315 fjz1 = _mm_macc_ps(dz11,fscal,fjz1);
317 /**************************
318 * CALCULATE INTERACTIONS *
319 **************************/
321 r12 = _mm_mul_ps(rsq12,rinv12);
323 /* Calculate table index by multiplying r with table scale and truncate to integer */
324 rt = _mm_mul_ps(r12,vftabscale);
325 vfitab = _mm_cvttps_epi32(rt);
327 vfeps = _mm_frcz_ps(rt);
329 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
331 twovfeps = _mm_add_ps(vfeps,vfeps);
332 vfitab = _mm_slli_epi32(vfitab,2);
334 /* CUBIC SPLINE TABLE ELECTROSTATICS */
335 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
336 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
337 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
338 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
339 _MM_TRANSPOSE4_PS(Y,F,G,H);
340 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
341 VV = _mm_macc_ps(vfeps,Fp,Y);
342 velec = _mm_mul_ps(qq12,VV);
343 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
344 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq12,FF),_mm_mul_ps(vftabscale,rinv12)));
346 /* Update potential sum for this i atom from the interaction with this j atom. */
347 velecsum = _mm_add_ps(velecsum,velec);
351 /* Update vectorial force */
352 fix1 = _mm_macc_ps(dx12,fscal,fix1);
353 fiy1 = _mm_macc_ps(dy12,fscal,fiy1);
354 fiz1 = _mm_macc_ps(dz12,fscal,fiz1);
356 fjx2 = _mm_macc_ps(dx12,fscal,fjx2);
357 fjy2 = _mm_macc_ps(dy12,fscal,fjy2);
358 fjz2 = _mm_macc_ps(dz12,fscal,fjz2);
360 /**************************
361 * CALCULATE INTERACTIONS *
362 **************************/
364 r13 = _mm_mul_ps(rsq13,rinv13);
366 /* Calculate table index by multiplying r with table scale and truncate to integer */
367 rt = _mm_mul_ps(r13,vftabscale);
368 vfitab = _mm_cvttps_epi32(rt);
370 vfeps = _mm_frcz_ps(rt);
372 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
374 twovfeps = _mm_add_ps(vfeps,vfeps);
375 vfitab = _mm_slli_epi32(vfitab,2);
377 /* CUBIC SPLINE TABLE ELECTROSTATICS */
378 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
379 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
380 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
381 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
382 _MM_TRANSPOSE4_PS(Y,F,G,H);
383 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
384 VV = _mm_macc_ps(vfeps,Fp,Y);
385 velec = _mm_mul_ps(qq13,VV);
386 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
387 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq13,FF),_mm_mul_ps(vftabscale,rinv13)));
389 /* Update potential sum for this i atom from the interaction with this j atom. */
390 velecsum = _mm_add_ps(velecsum,velec);
394 /* Update vectorial force */
395 fix1 = _mm_macc_ps(dx13,fscal,fix1);
396 fiy1 = _mm_macc_ps(dy13,fscal,fiy1);
397 fiz1 = _mm_macc_ps(dz13,fscal,fiz1);
399 fjx3 = _mm_macc_ps(dx13,fscal,fjx3);
400 fjy3 = _mm_macc_ps(dy13,fscal,fjy3);
401 fjz3 = _mm_macc_ps(dz13,fscal,fjz3);
403 /**************************
404 * CALCULATE INTERACTIONS *
405 **************************/
407 r21 = _mm_mul_ps(rsq21,rinv21);
409 /* Calculate table index by multiplying r with table scale and truncate to integer */
410 rt = _mm_mul_ps(r21,vftabscale);
411 vfitab = _mm_cvttps_epi32(rt);
413 vfeps = _mm_frcz_ps(rt);
415 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
417 twovfeps = _mm_add_ps(vfeps,vfeps);
418 vfitab = _mm_slli_epi32(vfitab,2);
420 /* CUBIC SPLINE TABLE ELECTROSTATICS */
421 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
422 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
423 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
424 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
425 _MM_TRANSPOSE4_PS(Y,F,G,H);
426 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
427 VV = _mm_macc_ps(vfeps,Fp,Y);
428 velec = _mm_mul_ps(qq21,VV);
429 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
430 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq21,FF),_mm_mul_ps(vftabscale,rinv21)));
432 /* Update potential sum for this i atom from the interaction with this j atom. */
433 velecsum = _mm_add_ps(velecsum,velec);
437 /* Update vectorial force */
438 fix2 = _mm_macc_ps(dx21,fscal,fix2);
439 fiy2 = _mm_macc_ps(dy21,fscal,fiy2);
440 fiz2 = _mm_macc_ps(dz21,fscal,fiz2);
442 fjx1 = _mm_macc_ps(dx21,fscal,fjx1);
443 fjy1 = _mm_macc_ps(dy21,fscal,fjy1);
444 fjz1 = _mm_macc_ps(dz21,fscal,fjz1);
446 /**************************
447 * CALCULATE INTERACTIONS *
448 **************************/
450 r22 = _mm_mul_ps(rsq22,rinv22);
452 /* Calculate table index by multiplying r with table scale and truncate to integer */
453 rt = _mm_mul_ps(r22,vftabscale);
454 vfitab = _mm_cvttps_epi32(rt);
456 vfeps = _mm_frcz_ps(rt);
458 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
460 twovfeps = _mm_add_ps(vfeps,vfeps);
461 vfitab = _mm_slli_epi32(vfitab,2);
463 /* CUBIC SPLINE TABLE ELECTROSTATICS */
464 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
465 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
466 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
467 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
468 _MM_TRANSPOSE4_PS(Y,F,G,H);
469 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
470 VV = _mm_macc_ps(vfeps,Fp,Y);
471 velec = _mm_mul_ps(qq22,VV);
472 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
473 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq22,FF),_mm_mul_ps(vftabscale,rinv22)));
475 /* Update potential sum for this i atom from the interaction with this j atom. */
476 velecsum = _mm_add_ps(velecsum,velec);
480 /* Update vectorial force */
481 fix2 = _mm_macc_ps(dx22,fscal,fix2);
482 fiy2 = _mm_macc_ps(dy22,fscal,fiy2);
483 fiz2 = _mm_macc_ps(dz22,fscal,fiz2);
485 fjx2 = _mm_macc_ps(dx22,fscal,fjx2);
486 fjy2 = _mm_macc_ps(dy22,fscal,fjy2);
487 fjz2 = _mm_macc_ps(dz22,fscal,fjz2);
489 /**************************
490 * CALCULATE INTERACTIONS *
491 **************************/
493 r23 = _mm_mul_ps(rsq23,rinv23);
495 /* Calculate table index by multiplying r with table scale and truncate to integer */
496 rt = _mm_mul_ps(r23,vftabscale);
497 vfitab = _mm_cvttps_epi32(rt);
499 vfeps = _mm_frcz_ps(rt);
501 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
503 twovfeps = _mm_add_ps(vfeps,vfeps);
504 vfitab = _mm_slli_epi32(vfitab,2);
506 /* CUBIC SPLINE TABLE ELECTROSTATICS */
507 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
508 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
509 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
510 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
511 _MM_TRANSPOSE4_PS(Y,F,G,H);
512 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
513 VV = _mm_macc_ps(vfeps,Fp,Y);
514 velec = _mm_mul_ps(qq23,VV);
515 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
516 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq23,FF),_mm_mul_ps(vftabscale,rinv23)));
518 /* Update potential sum for this i atom from the interaction with this j atom. */
519 velecsum = _mm_add_ps(velecsum,velec);
523 /* Update vectorial force */
524 fix2 = _mm_macc_ps(dx23,fscal,fix2);
525 fiy2 = _mm_macc_ps(dy23,fscal,fiy2);
526 fiz2 = _mm_macc_ps(dz23,fscal,fiz2);
528 fjx3 = _mm_macc_ps(dx23,fscal,fjx3);
529 fjy3 = _mm_macc_ps(dy23,fscal,fjy3);
530 fjz3 = _mm_macc_ps(dz23,fscal,fjz3);
532 /**************************
533 * CALCULATE INTERACTIONS *
534 **************************/
536 r31 = _mm_mul_ps(rsq31,rinv31);
538 /* Calculate table index by multiplying r with table scale and truncate to integer */
539 rt = _mm_mul_ps(r31,vftabscale);
540 vfitab = _mm_cvttps_epi32(rt);
542 vfeps = _mm_frcz_ps(rt);
544 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
546 twovfeps = _mm_add_ps(vfeps,vfeps);
547 vfitab = _mm_slli_epi32(vfitab,2);
549 /* CUBIC SPLINE TABLE ELECTROSTATICS */
550 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
551 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
552 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
553 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
554 _MM_TRANSPOSE4_PS(Y,F,G,H);
555 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
556 VV = _mm_macc_ps(vfeps,Fp,Y);
557 velec = _mm_mul_ps(qq31,VV);
558 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
559 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq31,FF),_mm_mul_ps(vftabscale,rinv31)));
561 /* Update potential sum for this i atom from the interaction with this j atom. */
562 velecsum = _mm_add_ps(velecsum,velec);
566 /* Update vectorial force */
567 fix3 = _mm_macc_ps(dx31,fscal,fix3);
568 fiy3 = _mm_macc_ps(dy31,fscal,fiy3);
569 fiz3 = _mm_macc_ps(dz31,fscal,fiz3);
571 fjx1 = _mm_macc_ps(dx31,fscal,fjx1);
572 fjy1 = _mm_macc_ps(dy31,fscal,fjy1);
573 fjz1 = _mm_macc_ps(dz31,fscal,fjz1);
575 /**************************
576 * CALCULATE INTERACTIONS *
577 **************************/
579 r32 = _mm_mul_ps(rsq32,rinv32);
581 /* Calculate table index by multiplying r with table scale and truncate to integer */
582 rt = _mm_mul_ps(r32,vftabscale);
583 vfitab = _mm_cvttps_epi32(rt);
585 vfeps = _mm_frcz_ps(rt);
587 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
589 twovfeps = _mm_add_ps(vfeps,vfeps);
590 vfitab = _mm_slli_epi32(vfitab,2);
592 /* CUBIC SPLINE TABLE ELECTROSTATICS */
593 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
594 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
595 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
596 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
597 _MM_TRANSPOSE4_PS(Y,F,G,H);
598 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
599 VV = _mm_macc_ps(vfeps,Fp,Y);
600 velec = _mm_mul_ps(qq32,VV);
601 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
602 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq32,FF),_mm_mul_ps(vftabscale,rinv32)));
604 /* Update potential sum for this i atom from the interaction with this j atom. */
605 velecsum = _mm_add_ps(velecsum,velec);
609 /* Update vectorial force */
610 fix3 = _mm_macc_ps(dx32,fscal,fix3);
611 fiy3 = _mm_macc_ps(dy32,fscal,fiy3);
612 fiz3 = _mm_macc_ps(dz32,fscal,fiz3);
614 fjx2 = _mm_macc_ps(dx32,fscal,fjx2);
615 fjy2 = _mm_macc_ps(dy32,fscal,fjy2);
616 fjz2 = _mm_macc_ps(dz32,fscal,fjz2);
618 /**************************
619 * CALCULATE INTERACTIONS *
620 **************************/
622 r33 = _mm_mul_ps(rsq33,rinv33);
624 /* Calculate table index by multiplying r with table scale and truncate to integer */
625 rt = _mm_mul_ps(r33,vftabscale);
626 vfitab = _mm_cvttps_epi32(rt);
628 vfeps = _mm_frcz_ps(rt);
630 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
632 twovfeps = _mm_add_ps(vfeps,vfeps);
633 vfitab = _mm_slli_epi32(vfitab,2);
635 /* CUBIC SPLINE TABLE ELECTROSTATICS */
636 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
637 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
638 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
639 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
640 _MM_TRANSPOSE4_PS(Y,F,G,H);
641 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
642 VV = _mm_macc_ps(vfeps,Fp,Y);
643 velec = _mm_mul_ps(qq33,VV);
644 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
645 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq33,FF),_mm_mul_ps(vftabscale,rinv33)));
647 /* Update potential sum for this i atom from the interaction with this j atom. */
648 velecsum = _mm_add_ps(velecsum,velec);
652 /* Update vectorial force */
653 fix3 = _mm_macc_ps(dx33,fscal,fix3);
654 fiy3 = _mm_macc_ps(dy33,fscal,fiy3);
655 fiz3 = _mm_macc_ps(dz33,fscal,fiz3);
657 fjx3 = _mm_macc_ps(dx33,fscal,fjx3);
658 fjy3 = _mm_macc_ps(dy33,fscal,fjy3);
659 fjz3 = _mm_macc_ps(dz33,fscal,fjz3);
661 fjptrA = f+j_coord_offsetA;
662 fjptrB = f+j_coord_offsetB;
663 fjptrC = f+j_coord_offsetC;
664 fjptrD = f+j_coord_offsetD;
666 gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
667 fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
669 /* Inner loop uses 414 flops */
675 /* Get j neighbor index, and coordinate index */
676 jnrlistA = jjnr[jidx];
677 jnrlistB = jjnr[jidx+1];
678 jnrlistC = jjnr[jidx+2];
679 jnrlistD = jjnr[jidx+3];
680 /* Sign of each element will be negative for non-real atoms.
681 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
682 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
684 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
685 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
686 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
687 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
688 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
689 j_coord_offsetA = DIM*jnrA;
690 j_coord_offsetB = DIM*jnrB;
691 j_coord_offsetC = DIM*jnrC;
692 j_coord_offsetD = DIM*jnrD;
694 /* load j atom coordinates */
695 gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
696 x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
697 &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
699 /* Calculate displacement vector */
700 dx11 = _mm_sub_ps(ix1,jx1);
701 dy11 = _mm_sub_ps(iy1,jy1);
702 dz11 = _mm_sub_ps(iz1,jz1);
703 dx12 = _mm_sub_ps(ix1,jx2);
704 dy12 = _mm_sub_ps(iy1,jy2);
705 dz12 = _mm_sub_ps(iz1,jz2);
706 dx13 = _mm_sub_ps(ix1,jx3);
707 dy13 = _mm_sub_ps(iy1,jy3);
708 dz13 = _mm_sub_ps(iz1,jz3);
709 dx21 = _mm_sub_ps(ix2,jx1);
710 dy21 = _mm_sub_ps(iy2,jy1);
711 dz21 = _mm_sub_ps(iz2,jz1);
712 dx22 = _mm_sub_ps(ix2,jx2);
713 dy22 = _mm_sub_ps(iy2,jy2);
714 dz22 = _mm_sub_ps(iz2,jz2);
715 dx23 = _mm_sub_ps(ix2,jx3);
716 dy23 = _mm_sub_ps(iy2,jy3);
717 dz23 = _mm_sub_ps(iz2,jz3);
718 dx31 = _mm_sub_ps(ix3,jx1);
719 dy31 = _mm_sub_ps(iy3,jy1);
720 dz31 = _mm_sub_ps(iz3,jz1);
721 dx32 = _mm_sub_ps(ix3,jx2);
722 dy32 = _mm_sub_ps(iy3,jy2);
723 dz32 = _mm_sub_ps(iz3,jz2);
724 dx33 = _mm_sub_ps(ix3,jx3);
725 dy33 = _mm_sub_ps(iy3,jy3);
726 dz33 = _mm_sub_ps(iz3,jz3);
728 /* Calculate squared distance and things based on it */
729 rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
730 rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
731 rsq13 = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
732 rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
733 rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
734 rsq23 = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
735 rsq31 = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
736 rsq32 = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
737 rsq33 = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
739 rinv11 = avx128fma_invsqrt_f(rsq11);
740 rinv12 = avx128fma_invsqrt_f(rsq12);
741 rinv13 = avx128fma_invsqrt_f(rsq13);
742 rinv21 = avx128fma_invsqrt_f(rsq21);
743 rinv22 = avx128fma_invsqrt_f(rsq22);
744 rinv23 = avx128fma_invsqrt_f(rsq23);
745 rinv31 = avx128fma_invsqrt_f(rsq31);
746 rinv32 = avx128fma_invsqrt_f(rsq32);
747 rinv33 = avx128fma_invsqrt_f(rsq33);
749 fjx1 = _mm_setzero_ps();
750 fjy1 = _mm_setzero_ps();
751 fjz1 = _mm_setzero_ps();
752 fjx2 = _mm_setzero_ps();
753 fjy2 = _mm_setzero_ps();
754 fjz2 = _mm_setzero_ps();
755 fjx3 = _mm_setzero_ps();
756 fjy3 = _mm_setzero_ps();
757 fjz3 = _mm_setzero_ps();
759 /**************************
760 * CALCULATE INTERACTIONS *
761 **************************/
763 r11 = _mm_mul_ps(rsq11,rinv11);
764 r11 = _mm_andnot_ps(dummy_mask,r11);
766 /* Calculate table index by multiplying r with table scale and truncate to integer */
767 rt = _mm_mul_ps(r11,vftabscale);
768 vfitab = _mm_cvttps_epi32(rt);
770 vfeps = _mm_frcz_ps(rt);
772 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
774 twovfeps = _mm_add_ps(vfeps,vfeps);
775 vfitab = _mm_slli_epi32(vfitab,2);
777 /* CUBIC SPLINE TABLE ELECTROSTATICS */
778 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
779 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
780 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
781 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
782 _MM_TRANSPOSE4_PS(Y,F,G,H);
783 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
784 VV = _mm_macc_ps(vfeps,Fp,Y);
785 velec = _mm_mul_ps(qq11,VV);
786 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
787 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq11,FF),_mm_mul_ps(vftabscale,rinv11)));
789 /* Update potential sum for this i atom from the interaction with this j atom. */
790 velec = _mm_andnot_ps(dummy_mask,velec);
791 velecsum = _mm_add_ps(velecsum,velec);
795 fscal = _mm_andnot_ps(dummy_mask,fscal);
797 /* Update vectorial force */
798 fix1 = _mm_macc_ps(dx11,fscal,fix1);
799 fiy1 = _mm_macc_ps(dy11,fscal,fiy1);
800 fiz1 = _mm_macc_ps(dz11,fscal,fiz1);
802 fjx1 = _mm_macc_ps(dx11,fscal,fjx1);
803 fjy1 = _mm_macc_ps(dy11,fscal,fjy1);
804 fjz1 = _mm_macc_ps(dz11,fscal,fjz1);
806 /**************************
807 * CALCULATE INTERACTIONS *
808 **************************/
810 r12 = _mm_mul_ps(rsq12,rinv12);
811 r12 = _mm_andnot_ps(dummy_mask,r12);
813 /* Calculate table index by multiplying r with table scale and truncate to integer */
814 rt = _mm_mul_ps(r12,vftabscale);
815 vfitab = _mm_cvttps_epi32(rt);
817 vfeps = _mm_frcz_ps(rt);
819 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
821 twovfeps = _mm_add_ps(vfeps,vfeps);
822 vfitab = _mm_slli_epi32(vfitab,2);
824 /* CUBIC SPLINE TABLE ELECTROSTATICS */
825 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
826 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
827 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
828 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
829 _MM_TRANSPOSE4_PS(Y,F,G,H);
830 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
831 VV = _mm_macc_ps(vfeps,Fp,Y);
832 velec = _mm_mul_ps(qq12,VV);
833 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
834 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq12,FF),_mm_mul_ps(vftabscale,rinv12)));
836 /* Update potential sum for this i atom from the interaction with this j atom. */
837 velec = _mm_andnot_ps(dummy_mask,velec);
838 velecsum = _mm_add_ps(velecsum,velec);
842 fscal = _mm_andnot_ps(dummy_mask,fscal);
844 /* Update vectorial force */
845 fix1 = _mm_macc_ps(dx12,fscal,fix1);
846 fiy1 = _mm_macc_ps(dy12,fscal,fiy1);
847 fiz1 = _mm_macc_ps(dz12,fscal,fiz1);
849 fjx2 = _mm_macc_ps(dx12,fscal,fjx2);
850 fjy2 = _mm_macc_ps(dy12,fscal,fjy2);
851 fjz2 = _mm_macc_ps(dz12,fscal,fjz2);
853 /**************************
854 * CALCULATE INTERACTIONS *
855 **************************/
857 r13 = _mm_mul_ps(rsq13,rinv13);
858 r13 = _mm_andnot_ps(dummy_mask,r13);
860 /* Calculate table index by multiplying r with table scale and truncate to integer */
861 rt = _mm_mul_ps(r13,vftabscale);
862 vfitab = _mm_cvttps_epi32(rt);
864 vfeps = _mm_frcz_ps(rt);
866 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
868 twovfeps = _mm_add_ps(vfeps,vfeps);
869 vfitab = _mm_slli_epi32(vfitab,2);
871 /* CUBIC SPLINE TABLE ELECTROSTATICS */
872 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
873 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
874 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
875 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
876 _MM_TRANSPOSE4_PS(Y,F,G,H);
877 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
878 VV = _mm_macc_ps(vfeps,Fp,Y);
879 velec = _mm_mul_ps(qq13,VV);
880 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
881 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq13,FF),_mm_mul_ps(vftabscale,rinv13)));
883 /* Update potential sum for this i atom from the interaction with this j atom. */
884 velec = _mm_andnot_ps(dummy_mask,velec);
885 velecsum = _mm_add_ps(velecsum,velec);
889 fscal = _mm_andnot_ps(dummy_mask,fscal);
891 /* Update vectorial force */
892 fix1 = _mm_macc_ps(dx13,fscal,fix1);
893 fiy1 = _mm_macc_ps(dy13,fscal,fiy1);
894 fiz1 = _mm_macc_ps(dz13,fscal,fiz1);
896 fjx3 = _mm_macc_ps(dx13,fscal,fjx3);
897 fjy3 = _mm_macc_ps(dy13,fscal,fjy3);
898 fjz3 = _mm_macc_ps(dz13,fscal,fjz3);
900 /**************************
901 * CALCULATE INTERACTIONS *
902 **************************/
904 r21 = _mm_mul_ps(rsq21,rinv21);
905 r21 = _mm_andnot_ps(dummy_mask,r21);
907 /* Calculate table index by multiplying r with table scale and truncate to integer */
908 rt = _mm_mul_ps(r21,vftabscale);
909 vfitab = _mm_cvttps_epi32(rt);
911 vfeps = _mm_frcz_ps(rt);
913 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
915 twovfeps = _mm_add_ps(vfeps,vfeps);
916 vfitab = _mm_slli_epi32(vfitab,2);
918 /* CUBIC SPLINE TABLE ELECTROSTATICS */
919 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
920 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
921 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
922 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
923 _MM_TRANSPOSE4_PS(Y,F,G,H);
924 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
925 VV = _mm_macc_ps(vfeps,Fp,Y);
926 velec = _mm_mul_ps(qq21,VV);
927 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
928 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq21,FF),_mm_mul_ps(vftabscale,rinv21)));
930 /* Update potential sum for this i atom from the interaction with this j atom. */
931 velec = _mm_andnot_ps(dummy_mask,velec);
932 velecsum = _mm_add_ps(velecsum,velec);
936 fscal = _mm_andnot_ps(dummy_mask,fscal);
938 /* Update vectorial force */
939 fix2 = _mm_macc_ps(dx21,fscal,fix2);
940 fiy2 = _mm_macc_ps(dy21,fscal,fiy2);
941 fiz2 = _mm_macc_ps(dz21,fscal,fiz2);
943 fjx1 = _mm_macc_ps(dx21,fscal,fjx1);
944 fjy1 = _mm_macc_ps(dy21,fscal,fjy1);
945 fjz1 = _mm_macc_ps(dz21,fscal,fjz1);
947 /**************************
948 * CALCULATE INTERACTIONS *
949 **************************/
951 r22 = _mm_mul_ps(rsq22,rinv22);
952 r22 = _mm_andnot_ps(dummy_mask,r22);
954 /* Calculate table index by multiplying r with table scale and truncate to integer */
955 rt = _mm_mul_ps(r22,vftabscale);
956 vfitab = _mm_cvttps_epi32(rt);
958 vfeps = _mm_frcz_ps(rt);
960 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
962 twovfeps = _mm_add_ps(vfeps,vfeps);
963 vfitab = _mm_slli_epi32(vfitab,2);
965 /* CUBIC SPLINE TABLE ELECTROSTATICS */
966 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
967 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
968 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
969 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
970 _MM_TRANSPOSE4_PS(Y,F,G,H);
971 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
972 VV = _mm_macc_ps(vfeps,Fp,Y);
973 velec = _mm_mul_ps(qq22,VV);
974 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
975 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq22,FF),_mm_mul_ps(vftabscale,rinv22)));
977 /* Update potential sum for this i atom from the interaction with this j atom. */
978 velec = _mm_andnot_ps(dummy_mask,velec);
979 velecsum = _mm_add_ps(velecsum,velec);
983 fscal = _mm_andnot_ps(dummy_mask,fscal);
985 /* Update vectorial force */
986 fix2 = _mm_macc_ps(dx22,fscal,fix2);
987 fiy2 = _mm_macc_ps(dy22,fscal,fiy2);
988 fiz2 = _mm_macc_ps(dz22,fscal,fiz2);
990 fjx2 = _mm_macc_ps(dx22,fscal,fjx2);
991 fjy2 = _mm_macc_ps(dy22,fscal,fjy2);
992 fjz2 = _mm_macc_ps(dz22,fscal,fjz2);
994 /**************************
995 * CALCULATE INTERACTIONS *
996 **************************/
998 r23 = _mm_mul_ps(rsq23,rinv23);
999 r23 = _mm_andnot_ps(dummy_mask,r23);
1001 /* Calculate table index by multiplying r with table scale and truncate to integer */
1002 rt = _mm_mul_ps(r23,vftabscale);
1003 vfitab = _mm_cvttps_epi32(rt);
1005 vfeps = _mm_frcz_ps(rt);
1007 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1009 twovfeps = _mm_add_ps(vfeps,vfeps);
1010 vfitab = _mm_slli_epi32(vfitab,2);
1012 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1013 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1014 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1015 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1016 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1017 _MM_TRANSPOSE4_PS(Y,F,G,H);
1018 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1019 VV = _mm_macc_ps(vfeps,Fp,Y);
1020 velec = _mm_mul_ps(qq23,VV);
1021 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1022 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq23,FF),_mm_mul_ps(vftabscale,rinv23)));
1024 /* Update potential sum for this i atom from the interaction with this j atom. */
1025 velec = _mm_andnot_ps(dummy_mask,velec);
1026 velecsum = _mm_add_ps(velecsum,velec);
1030 fscal = _mm_andnot_ps(dummy_mask,fscal);
1032 /* Update vectorial force */
1033 fix2 = _mm_macc_ps(dx23,fscal,fix2);
1034 fiy2 = _mm_macc_ps(dy23,fscal,fiy2);
1035 fiz2 = _mm_macc_ps(dz23,fscal,fiz2);
1037 fjx3 = _mm_macc_ps(dx23,fscal,fjx3);
1038 fjy3 = _mm_macc_ps(dy23,fscal,fjy3);
1039 fjz3 = _mm_macc_ps(dz23,fscal,fjz3);
1041 /**************************
1042 * CALCULATE INTERACTIONS *
1043 **************************/
1045 r31 = _mm_mul_ps(rsq31,rinv31);
1046 r31 = _mm_andnot_ps(dummy_mask,r31);
1048 /* Calculate table index by multiplying r with table scale and truncate to integer */
1049 rt = _mm_mul_ps(r31,vftabscale);
1050 vfitab = _mm_cvttps_epi32(rt);
1052 vfeps = _mm_frcz_ps(rt);
1054 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1056 twovfeps = _mm_add_ps(vfeps,vfeps);
1057 vfitab = _mm_slli_epi32(vfitab,2);
1059 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1060 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1061 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1062 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1063 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1064 _MM_TRANSPOSE4_PS(Y,F,G,H);
1065 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1066 VV = _mm_macc_ps(vfeps,Fp,Y);
1067 velec = _mm_mul_ps(qq31,VV);
1068 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1069 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq31,FF),_mm_mul_ps(vftabscale,rinv31)));
1071 /* Update potential sum for this i atom from the interaction with this j atom. */
1072 velec = _mm_andnot_ps(dummy_mask,velec);
1073 velecsum = _mm_add_ps(velecsum,velec);
1077 fscal = _mm_andnot_ps(dummy_mask,fscal);
1079 /* Update vectorial force */
1080 fix3 = _mm_macc_ps(dx31,fscal,fix3);
1081 fiy3 = _mm_macc_ps(dy31,fscal,fiy3);
1082 fiz3 = _mm_macc_ps(dz31,fscal,fiz3);
1084 fjx1 = _mm_macc_ps(dx31,fscal,fjx1);
1085 fjy1 = _mm_macc_ps(dy31,fscal,fjy1);
1086 fjz1 = _mm_macc_ps(dz31,fscal,fjz1);
1088 /**************************
1089 * CALCULATE INTERACTIONS *
1090 **************************/
1092 r32 = _mm_mul_ps(rsq32,rinv32);
1093 r32 = _mm_andnot_ps(dummy_mask,r32);
1095 /* Calculate table index by multiplying r with table scale and truncate to integer */
1096 rt = _mm_mul_ps(r32,vftabscale);
1097 vfitab = _mm_cvttps_epi32(rt);
1099 vfeps = _mm_frcz_ps(rt);
1101 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1103 twovfeps = _mm_add_ps(vfeps,vfeps);
1104 vfitab = _mm_slli_epi32(vfitab,2);
1106 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1107 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1108 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1109 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1110 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1111 _MM_TRANSPOSE4_PS(Y,F,G,H);
1112 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1113 VV = _mm_macc_ps(vfeps,Fp,Y);
1114 velec = _mm_mul_ps(qq32,VV);
1115 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1116 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq32,FF),_mm_mul_ps(vftabscale,rinv32)));
1118 /* Update potential sum for this i atom from the interaction with this j atom. */
1119 velec = _mm_andnot_ps(dummy_mask,velec);
1120 velecsum = _mm_add_ps(velecsum,velec);
1124 fscal = _mm_andnot_ps(dummy_mask,fscal);
1126 /* Update vectorial force */
1127 fix3 = _mm_macc_ps(dx32,fscal,fix3);
1128 fiy3 = _mm_macc_ps(dy32,fscal,fiy3);
1129 fiz3 = _mm_macc_ps(dz32,fscal,fiz3);
1131 fjx2 = _mm_macc_ps(dx32,fscal,fjx2);
1132 fjy2 = _mm_macc_ps(dy32,fscal,fjy2);
1133 fjz2 = _mm_macc_ps(dz32,fscal,fjz2);
1135 /**************************
1136 * CALCULATE INTERACTIONS *
1137 **************************/
1139 r33 = _mm_mul_ps(rsq33,rinv33);
1140 r33 = _mm_andnot_ps(dummy_mask,r33);
1142 /* Calculate table index by multiplying r with table scale and truncate to integer */
1143 rt = _mm_mul_ps(r33,vftabscale);
1144 vfitab = _mm_cvttps_epi32(rt);
1146 vfeps = _mm_frcz_ps(rt);
1148 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1150 twovfeps = _mm_add_ps(vfeps,vfeps);
1151 vfitab = _mm_slli_epi32(vfitab,2);
1153 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1154 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1155 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1156 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1157 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1158 _MM_TRANSPOSE4_PS(Y,F,G,H);
1159 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1160 VV = _mm_macc_ps(vfeps,Fp,Y);
1161 velec = _mm_mul_ps(qq33,VV);
1162 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1163 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq33,FF),_mm_mul_ps(vftabscale,rinv33)));
1165 /* Update potential sum for this i atom from the interaction with this j atom. */
1166 velec = _mm_andnot_ps(dummy_mask,velec);
1167 velecsum = _mm_add_ps(velecsum,velec);
1171 fscal = _mm_andnot_ps(dummy_mask,fscal);
1173 /* Update vectorial force */
1174 fix3 = _mm_macc_ps(dx33,fscal,fix3);
1175 fiy3 = _mm_macc_ps(dy33,fscal,fiy3);
1176 fiz3 = _mm_macc_ps(dz33,fscal,fiz3);
1178 fjx3 = _mm_macc_ps(dx33,fscal,fjx3);
1179 fjy3 = _mm_macc_ps(dy33,fscal,fjy3);
1180 fjz3 = _mm_macc_ps(dz33,fscal,fjz3);
1182 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1183 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1184 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1185 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1187 gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
1188 fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1190 /* Inner loop uses 423 flops */
1193 /* End of innermost loop */
1195 gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1196 f+i_coord_offset+DIM,fshift+i_shift_offset);
1199 /* Update potential energies */
1200 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1202 /* Increment number of inner iterations */
1203 inneriter += j_index_end - j_index_start;
1205 /* Outer loop uses 19 flops */
1208 /* Increment number of outer iterations */
1211 /* Update outer/inner flops */
1213 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_VF,outeriter*19 + inneriter*423);
1216 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwNone_GeomW4W4_F_avx_128_fma_single
1217 * Electrostatics interaction: CubicSplineTable
1218 * VdW interaction: None
1219 * Geometry: Water4-Water4
1220 * Calculate force/pot: Force
1223 nb_kernel_ElecCSTab_VdwNone_GeomW4W4_F_avx_128_fma_single
1224 (t_nblist * gmx_restrict nlist,
1225 rvec * gmx_restrict xx,
1226 rvec * gmx_restrict ff,
1227 struct t_forcerec * gmx_restrict fr,
1228 t_mdatoms * gmx_restrict mdatoms,
1229 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1230 t_nrnb * gmx_restrict nrnb)
1232 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1233 * just 0 for non-waters.
1234 * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
1235 * jnr indices corresponding to data put in the four positions in the SIMD register.
1237 int i_shift_offset,i_coord_offset,outeriter,inneriter;
1238 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1239 int jnrA,jnrB,jnrC,jnrD;
1240 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1241 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1242 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
1243 real rcutoff_scalar;
1244 real *shiftvec,*fshift,*x,*f;
1245 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1246 real scratch[4*DIM];
1247 __m128 fscal,rcutoff,rcutoff2,jidxall;
1249 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1251 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1253 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1254 int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1255 __m128 jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1256 int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1257 __m128 jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1258 int vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
1259 __m128 jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1260 __m128 dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1261 __m128 dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1262 __m128 dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1263 __m128 dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1264 __m128 dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1265 __m128 dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1266 __m128 dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1267 __m128 dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1268 __m128 dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1269 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
1272 __m128i ifour = _mm_set1_epi32(4);
1273 __m128 rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
1275 __m128 dummy_mask,cutoff_mask;
1276 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
1277 __m128 one = _mm_set1_ps(1.0);
1278 __m128 two = _mm_set1_ps(2.0);
1284 jindex = nlist->jindex;
1286 shiftidx = nlist->shift;
1288 shiftvec = fr->shift_vec[0];
1289 fshift = fr->fshift[0];
1290 facel = _mm_set1_ps(fr->ic->epsfac);
1291 charge = mdatoms->chargeA;
1293 vftab = kernel_data->table_elec->data;
1294 vftabscale = _mm_set1_ps(kernel_data->table_elec->scale);
1296 /* Setup water-specific parameters */
1297 inr = nlist->iinr[0];
1298 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
1299 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
1300 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
1302 jq1 = _mm_set1_ps(charge[inr+1]);
1303 jq2 = _mm_set1_ps(charge[inr+2]);
1304 jq3 = _mm_set1_ps(charge[inr+3]);
1305 qq11 = _mm_mul_ps(iq1,jq1);
1306 qq12 = _mm_mul_ps(iq1,jq2);
1307 qq13 = _mm_mul_ps(iq1,jq3);
1308 qq21 = _mm_mul_ps(iq2,jq1);
1309 qq22 = _mm_mul_ps(iq2,jq2);
1310 qq23 = _mm_mul_ps(iq2,jq3);
1311 qq31 = _mm_mul_ps(iq3,jq1);
1312 qq32 = _mm_mul_ps(iq3,jq2);
1313 qq33 = _mm_mul_ps(iq3,jq3);
1315 /* Avoid stupid compiler warnings */
1316 jnrA = jnrB = jnrC = jnrD = 0;
1317 j_coord_offsetA = 0;
1318 j_coord_offsetB = 0;
1319 j_coord_offsetC = 0;
1320 j_coord_offsetD = 0;
1325 for(iidx=0;iidx<4*DIM;iidx++)
1327 scratch[iidx] = 0.0;
1330 /* Start outer loop over neighborlists */
1331 for(iidx=0; iidx<nri; iidx++)
1333 /* Load shift vector for this list */
1334 i_shift_offset = DIM*shiftidx[iidx];
1336 /* Load limits for loop over neighbors */
1337 j_index_start = jindex[iidx];
1338 j_index_end = jindex[iidx+1];
1340 /* Get outer coordinate index */
1342 i_coord_offset = DIM*inr;
1344 /* Load i particle coords and add shift vector */
1345 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
1346 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1348 fix1 = _mm_setzero_ps();
1349 fiy1 = _mm_setzero_ps();
1350 fiz1 = _mm_setzero_ps();
1351 fix2 = _mm_setzero_ps();
1352 fiy2 = _mm_setzero_ps();
1353 fiz2 = _mm_setzero_ps();
1354 fix3 = _mm_setzero_ps();
1355 fiy3 = _mm_setzero_ps();
1356 fiz3 = _mm_setzero_ps();
1358 /* Start inner kernel loop */
1359 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1362 /* Get j neighbor index, and coordinate index */
1364 jnrB = jjnr[jidx+1];
1365 jnrC = jjnr[jidx+2];
1366 jnrD = jjnr[jidx+3];
1367 j_coord_offsetA = DIM*jnrA;
1368 j_coord_offsetB = DIM*jnrB;
1369 j_coord_offsetC = DIM*jnrC;
1370 j_coord_offsetD = DIM*jnrD;
1372 /* load j atom coordinates */
1373 gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
1374 x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
1375 &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1377 /* Calculate displacement vector */
1378 dx11 = _mm_sub_ps(ix1,jx1);
1379 dy11 = _mm_sub_ps(iy1,jy1);
1380 dz11 = _mm_sub_ps(iz1,jz1);
1381 dx12 = _mm_sub_ps(ix1,jx2);
1382 dy12 = _mm_sub_ps(iy1,jy2);
1383 dz12 = _mm_sub_ps(iz1,jz2);
1384 dx13 = _mm_sub_ps(ix1,jx3);
1385 dy13 = _mm_sub_ps(iy1,jy3);
1386 dz13 = _mm_sub_ps(iz1,jz3);
1387 dx21 = _mm_sub_ps(ix2,jx1);
1388 dy21 = _mm_sub_ps(iy2,jy1);
1389 dz21 = _mm_sub_ps(iz2,jz1);
1390 dx22 = _mm_sub_ps(ix2,jx2);
1391 dy22 = _mm_sub_ps(iy2,jy2);
1392 dz22 = _mm_sub_ps(iz2,jz2);
1393 dx23 = _mm_sub_ps(ix2,jx3);
1394 dy23 = _mm_sub_ps(iy2,jy3);
1395 dz23 = _mm_sub_ps(iz2,jz3);
1396 dx31 = _mm_sub_ps(ix3,jx1);
1397 dy31 = _mm_sub_ps(iy3,jy1);
1398 dz31 = _mm_sub_ps(iz3,jz1);
1399 dx32 = _mm_sub_ps(ix3,jx2);
1400 dy32 = _mm_sub_ps(iy3,jy2);
1401 dz32 = _mm_sub_ps(iz3,jz2);
1402 dx33 = _mm_sub_ps(ix3,jx3);
1403 dy33 = _mm_sub_ps(iy3,jy3);
1404 dz33 = _mm_sub_ps(iz3,jz3);
1406 /* Calculate squared distance and things based on it */
1407 rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1408 rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1409 rsq13 = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
1410 rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1411 rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1412 rsq23 = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
1413 rsq31 = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
1414 rsq32 = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
1415 rsq33 = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
1417 rinv11 = avx128fma_invsqrt_f(rsq11);
1418 rinv12 = avx128fma_invsqrt_f(rsq12);
1419 rinv13 = avx128fma_invsqrt_f(rsq13);
1420 rinv21 = avx128fma_invsqrt_f(rsq21);
1421 rinv22 = avx128fma_invsqrt_f(rsq22);
1422 rinv23 = avx128fma_invsqrt_f(rsq23);
1423 rinv31 = avx128fma_invsqrt_f(rsq31);
1424 rinv32 = avx128fma_invsqrt_f(rsq32);
1425 rinv33 = avx128fma_invsqrt_f(rsq33);
1427 fjx1 = _mm_setzero_ps();
1428 fjy1 = _mm_setzero_ps();
1429 fjz1 = _mm_setzero_ps();
1430 fjx2 = _mm_setzero_ps();
1431 fjy2 = _mm_setzero_ps();
1432 fjz2 = _mm_setzero_ps();
1433 fjx3 = _mm_setzero_ps();
1434 fjy3 = _mm_setzero_ps();
1435 fjz3 = _mm_setzero_ps();
1437 /**************************
1438 * CALCULATE INTERACTIONS *
1439 **************************/
1441 r11 = _mm_mul_ps(rsq11,rinv11);
1443 /* Calculate table index by multiplying r with table scale and truncate to integer */
1444 rt = _mm_mul_ps(r11,vftabscale);
1445 vfitab = _mm_cvttps_epi32(rt);
1447 vfeps = _mm_frcz_ps(rt);
1449 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1451 twovfeps = _mm_add_ps(vfeps,vfeps);
1452 vfitab = _mm_slli_epi32(vfitab,2);
1454 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1455 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1456 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1457 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1458 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1459 _MM_TRANSPOSE4_PS(Y,F,G,H);
1460 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1461 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1462 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq11,FF),_mm_mul_ps(vftabscale,rinv11)));
1466 /* Update vectorial force */
1467 fix1 = _mm_macc_ps(dx11,fscal,fix1);
1468 fiy1 = _mm_macc_ps(dy11,fscal,fiy1);
1469 fiz1 = _mm_macc_ps(dz11,fscal,fiz1);
1471 fjx1 = _mm_macc_ps(dx11,fscal,fjx1);
1472 fjy1 = _mm_macc_ps(dy11,fscal,fjy1);
1473 fjz1 = _mm_macc_ps(dz11,fscal,fjz1);
1475 /**************************
1476 * CALCULATE INTERACTIONS *
1477 **************************/
1479 r12 = _mm_mul_ps(rsq12,rinv12);
1481 /* Calculate table index by multiplying r with table scale and truncate to integer */
1482 rt = _mm_mul_ps(r12,vftabscale);
1483 vfitab = _mm_cvttps_epi32(rt);
1485 vfeps = _mm_frcz_ps(rt);
1487 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1489 twovfeps = _mm_add_ps(vfeps,vfeps);
1490 vfitab = _mm_slli_epi32(vfitab,2);
1492 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1493 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1494 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1495 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1496 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1497 _MM_TRANSPOSE4_PS(Y,F,G,H);
1498 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1499 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1500 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq12,FF),_mm_mul_ps(vftabscale,rinv12)));
1504 /* Update vectorial force */
1505 fix1 = _mm_macc_ps(dx12,fscal,fix1);
1506 fiy1 = _mm_macc_ps(dy12,fscal,fiy1);
1507 fiz1 = _mm_macc_ps(dz12,fscal,fiz1);
1509 fjx2 = _mm_macc_ps(dx12,fscal,fjx2);
1510 fjy2 = _mm_macc_ps(dy12,fscal,fjy2);
1511 fjz2 = _mm_macc_ps(dz12,fscal,fjz2);
1513 /**************************
1514 * CALCULATE INTERACTIONS *
1515 **************************/
1517 r13 = _mm_mul_ps(rsq13,rinv13);
1519 /* Calculate table index by multiplying r with table scale and truncate to integer */
1520 rt = _mm_mul_ps(r13,vftabscale);
1521 vfitab = _mm_cvttps_epi32(rt);
1523 vfeps = _mm_frcz_ps(rt);
1525 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1527 twovfeps = _mm_add_ps(vfeps,vfeps);
1528 vfitab = _mm_slli_epi32(vfitab,2);
1530 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1531 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1532 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1533 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1534 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1535 _MM_TRANSPOSE4_PS(Y,F,G,H);
1536 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1537 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1538 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq13,FF),_mm_mul_ps(vftabscale,rinv13)));
1542 /* Update vectorial force */
1543 fix1 = _mm_macc_ps(dx13,fscal,fix1);
1544 fiy1 = _mm_macc_ps(dy13,fscal,fiy1);
1545 fiz1 = _mm_macc_ps(dz13,fscal,fiz1);
1547 fjx3 = _mm_macc_ps(dx13,fscal,fjx3);
1548 fjy3 = _mm_macc_ps(dy13,fscal,fjy3);
1549 fjz3 = _mm_macc_ps(dz13,fscal,fjz3);
1551 /**************************
1552 * CALCULATE INTERACTIONS *
1553 **************************/
1555 r21 = _mm_mul_ps(rsq21,rinv21);
1557 /* Calculate table index by multiplying r with table scale and truncate to integer */
1558 rt = _mm_mul_ps(r21,vftabscale);
1559 vfitab = _mm_cvttps_epi32(rt);
1561 vfeps = _mm_frcz_ps(rt);
1563 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1565 twovfeps = _mm_add_ps(vfeps,vfeps);
1566 vfitab = _mm_slli_epi32(vfitab,2);
1568 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1569 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1570 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1571 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1572 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1573 _MM_TRANSPOSE4_PS(Y,F,G,H);
1574 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1575 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1576 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq21,FF),_mm_mul_ps(vftabscale,rinv21)));
1580 /* Update vectorial force */
1581 fix2 = _mm_macc_ps(dx21,fscal,fix2);
1582 fiy2 = _mm_macc_ps(dy21,fscal,fiy2);
1583 fiz2 = _mm_macc_ps(dz21,fscal,fiz2);
1585 fjx1 = _mm_macc_ps(dx21,fscal,fjx1);
1586 fjy1 = _mm_macc_ps(dy21,fscal,fjy1);
1587 fjz1 = _mm_macc_ps(dz21,fscal,fjz1);
1589 /**************************
1590 * CALCULATE INTERACTIONS *
1591 **************************/
1593 r22 = _mm_mul_ps(rsq22,rinv22);
1595 /* Calculate table index by multiplying r with table scale and truncate to integer */
1596 rt = _mm_mul_ps(r22,vftabscale);
1597 vfitab = _mm_cvttps_epi32(rt);
1599 vfeps = _mm_frcz_ps(rt);
1601 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1603 twovfeps = _mm_add_ps(vfeps,vfeps);
1604 vfitab = _mm_slli_epi32(vfitab,2);
1606 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1607 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1608 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1609 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1610 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1611 _MM_TRANSPOSE4_PS(Y,F,G,H);
1612 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1613 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1614 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq22,FF),_mm_mul_ps(vftabscale,rinv22)));
1618 /* Update vectorial force */
1619 fix2 = _mm_macc_ps(dx22,fscal,fix2);
1620 fiy2 = _mm_macc_ps(dy22,fscal,fiy2);
1621 fiz2 = _mm_macc_ps(dz22,fscal,fiz2);
1623 fjx2 = _mm_macc_ps(dx22,fscal,fjx2);
1624 fjy2 = _mm_macc_ps(dy22,fscal,fjy2);
1625 fjz2 = _mm_macc_ps(dz22,fscal,fjz2);
1627 /**************************
1628 * CALCULATE INTERACTIONS *
1629 **************************/
1631 r23 = _mm_mul_ps(rsq23,rinv23);
1633 /* Calculate table index by multiplying r with table scale and truncate to integer */
1634 rt = _mm_mul_ps(r23,vftabscale);
1635 vfitab = _mm_cvttps_epi32(rt);
1637 vfeps = _mm_frcz_ps(rt);
1639 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1641 twovfeps = _mm_add_ps(vfeps,vfeps);
1642 vfitab = _mm_slli_epi32(vfitab,2);
1644 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1645 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1646 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1647 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1648 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1649 _MM_TRANSPOSE4_PS(Y,F,G,H);
1650 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1651 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1652 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq23,FF),_mm_mul_ps(vftabscale,rinv23)));
1656 /* Update vectorial force */
1657 fix2 = _mm_macc_ps(dx23,fscal,fix2);
1658 fiy2 = _mm_macc_ps(dy23,fscal,fiy2);
1659 fiz2 = _mm_macc_ps(dz23,fscal,fiz2);
1661 fjx3 = _mm_macc_ps(dx23,fscal,fjx3);
1662 fjy3 = _mm_macc_ps(dy23,fscal,fjy3);
1663 fjz3 = _mm_macc_ps(dz23,fscal,fjz3);
1665 /**************************
1666 * CALCULATE INTERACTIONS *
1667 **************************/
1669 r31 = _mm_mul_ps(rsq31,rinv31);
1671 /* Calculate table index by multiplying r with table scale and truncate to integer */
1672 rt = _mm_mul_ps(r31,vftabscale);
1673 vfitab = _mm_cvttps_epi32(rt);
1675 vfeps = _mm_frcz_ps(rt);
1677 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1679 twovfeps = _mm_add_ps(vfeps,vfeps);
1680 vfitab = _mm_slli_epi32(vfitab,2);
1682 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1683 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1684 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1685 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1686 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1687 _MM_TRANSPOSE4_PS(Y,F,G,H);
1688 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1689 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1690 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq31,FF),_mm_mul_ps(vftabscale,rinv31)));
1694 /* Update vectorial force */
1695 fix3 = _mm_macc_ps(dx31,fscal,fix3);
1696 fiy3 = _mm_macc_ps(dy31,fscal,fiy3);
1697 fiz3 = _mm_macc_ps(dz31,fscal,fiz3);
1699 fjx1 = _mm_macc_ps(dx31,fscal,fjx1);
1700 fjy1 = _mm_macc_ps(dy31,fscal,fjy1);
1701 fjz1 = _mm_macc_ps(dz31,fscal,fjz1);
1703 /**************************
1704 * CALCULATE INTERACTIONS *
1705 **************************/
1707 r32 = _mm_mul_ps(rsq32,rinv32);
1709 /* Calculate table index by multiplying r with table scale and truncate to integer */
1710 rt = _mm_mul_ps(r32,vftabscale);
1711 vfitab = _mm_cvttps_epi32(rt);
1713 vfeps = _mm_frcz_ps(rt);
1715 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1717 twovfeps = _mm_add_ps(vfeps,vfeps);
1718 vfitab = _mm_slli_epi32(vfitab,2);
1720 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1721 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1722 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1723 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1724 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1725 _MM_TRANSPOSE4_PS(Y,F,G,H);
1726 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1727 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1728 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq32,FF),_mm_mul_ps(vftabscale,rinv32)));
1732 /* Update vectorial force */
1733 fix3 = _mm_macc_ps(dx32,fscal,fix3);
1734 fiy3 = _mm_macc_ps(dy32,fscal,fiy3);
1735 fiz3 = _mm_macc_ps(dz32,fscal,fiz3);
1737 fjx2 = _mm_macc_ps(dx32,fscal,fjx2);
1738 fjy2 = _mm_macc_ps(dy32,fscal,fjy2);
1739 fjz2 = _mm_macc_ps(dz32,fscal,fjz2);
1741 /**************************
1742 * CALCULATE INTERACTIONS *
1743 **************************/
1745 r33 = _mm_mul_ps(rsq33,rinv33);
1747 /* Calculate table index by multiplying r with table scale and truncate to integer */
1748 rt = _mm_mul_ps(r33,vftabscale);
1749 vfitab = _mm_cvttps_epi32(rt);
1751 vfeps = _mm_frcz_ps(rt);
1753 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1755 twovfeps = _mm_add_ps(vfeps,vfeps);
1756 vfitab = _mm_slli_epi32(vfitab,2);
1758 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1759 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1760 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1761 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1762 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1763 _MM_TRANSPOSE4_PS(Y,F,G,H);
1764 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1765 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1766 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq33,FF),_mm_mul_ps(vftabscale,rinv33)));
1770 /* Update vectorial force */
1771 fix3 = _mm_macc_ps(dx33,fscal,fix3);
1772 fiy3 = _mm_macc_ps(dy33,fscal,fiy3);
1773 fiz3 = _mm_macc_ps(dz33,fscal,fiz3);
1775 fjx3 = _mm_macc_ps(dx33,fscal,fjx3);
1776 fjy3 = _mm_macc_ps(dy33,fscal,fjy3);
1777 fjz3 = _mm_macc_ps(dz33,fscal,fjz3);
1779 fjptrA = f+j_coord_offsetA;
1780 fjptrB = f+j_coord_offsetB;
1781 fjptrC = f+j_coord_offsetC;
1782 fjptrD = f+j_coord_offsetD;
1784 gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
1785 fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1787 /* Inner loop uses 378 flops */
1790 if(jidx<j_index_end)
1793 /* Get j neighbor index, and coordinate index */
1794 jnrlistA = jjnr[jidx];
1795 jnrlistB = jjnr[jidx+1];
1796 jnrlistC = jjnr[jidx+2];
1797 jnrlistD = jjnr[jidx+3];
1798 /* Sign of each element will be negative for non-real atoms.
1799 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1800 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1802 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1803 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
1804 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
1805 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
1806 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
1807 j_coord_offsetA = DIM*jnrA;
1808 j_coord_offsetB = DIM*jnrB;
1809 j_coord_offsetC = DIM*jnrC;
1810 j_coord_offsetD = DIM*jnrD;
1812 /* load j atom coordinates */
1813 gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
1814 x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
1815 &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1817 /* Calculate displacement vector */
1818 dx11 = _mm_sub_ps(ix1,jx1);
1819 dy11 = _mm_sub_ps(iy1,jy1);
1820 dz11 = _mm_sub_ps(iz1,jz1);
1821 dx12 = _mm_sub_ps(ix1,jx2);
1822 dy12 = _mm_sub_ps(iy1,jy2);
1823 dz12 = _mm_sub_ps(iz1,jz2);
1824 dx13 = _mm_sub_ps(ix1,jx3);
1825 dy13 = _mm_sub_ps(iy1,jy3);
1826 dz13 = _mm_sub_ps(iz1,jz3);
1827 dx21 = _mm_sub_ps(ix2,jx1);
1828 dy21 = _mm_sub_ps(iy2,jy1);
1829 dz21 = _mm_sub_ps(iz2,jz1);
1830 dx22 = _mm_sub_ps(ix2,jx2);
1831 dy22 = _mm_sub_ps(iy2,jy2);
1832 dz22 = _mm_sub_ps(iz2,jz2);
1833 dx23 = _mm_sub_ps(ix2,jx3);
1834 dy23 = _mm_sub_ps(iy2,jy3);
1835 dz23 = _mm_sub_ps(iz2,jz3);
1836 dx31 = _mm_sub_ps(ix3,jx1);
1837 dy31 = _mm_sub_ps(iy3,jy1);
1838 dz31 = _mm_sub_ps(iz3,jz1);
1839 dx32 = _mm_sub_ps(ix3,jx2);
1840 dy32 = _mm_sub_ps(iy3,jy2);
1841 dz32 = _mm_sub_ps(iz3,jz2);
1842 dx33 = _mm_sub_ps(ix3,jx3);
1843 dy33 = _mm_sub_ps(iy3,jy3);
1844 dz33 = _mm_sub_ps(iz3,jz3);
1846 /* Calculate squared distance and things based on it */
1847 rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1848 rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1849 rsq13 = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
1850 rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1851 rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1852 rsq23 = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
1853 rsq31 = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
1854 rsq32 = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
1855 rsq33 = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
1857 rinv11 = avx128fma_invsqrt_f(rsq11);
1858 rinv12 = avx128fma_invsqrt_f(rsq12);
1859 rinv13 = avx128fma_invsqrt_f(rsq13);
1860 rinv21 = avx128fma_invsqrt_f(rsq21);
1861 rinv22 = avx128fma_invsqrt_f(rsq22);
1862 rinv23 = avx128fma_invsqrt_f(rsq23);
1863 rinv31 = avx128fma_invsqrt_f(rsq31);
1864 rinv32 = avx128fma_invsqrt_f(rsq32);
1865 rinv33 = avx128fma_invsqrt_f(rsq33);
1867 fjx1 = _mm_setzero_ps();
1868 fjy1 = _mm_setzero_ps();
1869 fjz1 = _mm_setzero_ps();
1870 fjx2 = _mm_setzero_ps();
1871 fjy2 = _mm_setzero_ps();
1872 fjz2 = _mm_setzero_ps();
1873 fjx3 = _mm_setzero_ps();
1874 fjy3 = _mm_setzero_ps();
1875 fjz3 = _mm_setzero_ps();
1877 /**************************
1878 * CALCULATE INTERACTIONS *
1879 **************************/
1881 r11 = _mm_mul_ps(rsq11,rinv11);
1882 r11 = _mm_andnot_ps(dummy_mask,r11);
1884 /* Calculate table index by multiplying r with table scale and truncate to integer */
1885 rt = _mm_mul_ps(r11,vftabscale);
1886 vfitab = _mm_cvttps_epi32(rt);
1888 vfeps = _mm_frcz_ps(rt);
1890 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1892 twovfeps = _mm_add_ps(vfeps,vfeps);
1893 vfitab = _mm_slli_epi32(vfitab,2);
1895 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1896 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1897 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1898 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1899 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1900 _MM_TRANSPOSE4_PS(Y,F,G,H);
1901 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1902 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1903 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq11,FF),_mm_mul_ps(vftabscale,rinv11)));
1907 fscal = _mm_andnot_ps(dummy_mask,fscal);
1909 /* Update vectorial force */
1910 fix1 = _mm_macc_ps(dx11,fscal,fix1);
1911 fiy1 = _mm_macc_ps(dy11,fscal,fiy1);
1912 fiz1 = _mm_macc_ps(dz11,fscal,fiz1);
1914 fjx1 = _mm_macc_ps(dx11,fscal,fjx1);
1915 fjy1 = _mm_macc_ps(dy11,fscal,fjy1);
1916 fjz1 = _mm_macc_ps(dz11,fscal,fjz1);
1918 /**************************
1919 * CALCULATE INTERACTIONS *
1920 **************************/
1922 r12 = _mm_mul_ps(rsq12,rinv12);
1923 r12 = _mm_andnot_ps(dummy_mask,r12);
1925 /* Calculate table index by multiplying r with table scale and truncate to integer */
1926 rt = _mm_mul_ps(r12,vftabscale);
1927 vfitab = _mm_cvttps_epi32(rt);
1929 vfeps = _mm_frcz_ps(rt);
1931 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1933 twovfeps = _mm_add_ps(vfeps,vfeps);
1934 vfitab = _mm_slli_epi32(vfitab,2);
1936 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1937 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1938 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1939 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1940 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1941 _MM_TRANSPOSE4_PS(Y,F,G,H);
1942 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1943 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1944 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq12,FF),_mm_mul_ps(vftabscale,rinv12)));
1948 fscal = _mm_andnot_ps(dummy_mask,fscal);
1950 /* Update vectorial force */
1951 fix1 = _mm_macc_ps(dx12,fscal,fix1);
1952 fiy1 = _mm_macc_ps(dy12,fscal,fiy1);
1953 fiz1 = _mm_macc_ps(dz12,fscal,fiz1);
1955 fjx2 = _mm_macc_ps(dx12,fscal,fjx2);
1956 fjy2 = _mm_macc_ps(dy12,fscal,fjy2);
1957 fjz2 = _mm_macc_ps(dz12,fscal,fjz2);
1959 /**************************
1960 * CALCULATE INTERACTIONS *
1961 **************************/
1963 r13 = _mm_mul_ps(rsq13,rinv13);
1964 r13 = _mm_andnot_ps(dummy_mask,r13);
1966 /* Calculate table index by multiplying r with table scale and truncate to integer */
1967 rt = _mm_mul_ps(r13,vftabscale);
1968 vfitab = _mm_cvttps_epi32(rt);
1970 vfeps = _mm_frcz_ps(rt);
1972 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1974 twovfeps = _mm_add_ps(vfeps,vfeps);
1975 vfitab = _mm_slli_epi32(vfitab,2);
1977 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1978 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1979 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1980 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1981 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1982 _MM_TRANSPOSE4_PS(Y,F,G,H);
1983 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1984 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1985 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq13,FF),_mm_mul_ps(vftabscale,rinv13)));
1989 fscal = _mm_andnot_ps(dummy_mask,fscal);
1991 /* Update vectorial force */
1992 fix1 = _mm_macc_ps(dx13,fscal,fix1);
1993 fiy1 = _mm_macc_ps(dy13,fscal,fiy1);
1994 fiz1 = _mm_macc_ps(dz13,fscal,fiz1);
1996 fjx3 = _mm_macc_ps(dx13,fscal,fjx3);
1997 fjy3 = _mm_macc_ps(dy13,fscal,fjy3);
1998 fjz3 = _mm_macc_ps(dz13,fscal,fjz3);
2000 /**************************
2001 * CALCULATE INTERACTIONS *
2002 **************************/
2004 r21 = _mm_mul_ps(rsq21,rinv21);
2005 r21 = _mm_andnot_ps(dummy_mask,r21);
2007 /* Calculate table index by multiplying r with table scale and truncate to integer */
2008 rt = _mm_mul_ps(r21,vftabscale);
2009 vfitab = _mm_cvttps_epi32(rt);
2011 vfeps = _mm_frcz_ps(rt);
2013 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
2015 twovfeps = _mm_add_ps(vfeps,vfeps);
2016 vfitab = _mm_slli_epi32(vfitab,2);
2018 /* CUBIC SPLINE TABLE ELECTROSTATICS */
2019 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
2020 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
2021 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
2022 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
2023 _MM_TRANSPOSE4_PS(Y,F,G,H);
2024 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
2025 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
2026 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq21,FF),_mm_mul_ps(vftabscale,rinv21)));
2030 fscal = _mm_andnot_ps(dummy_mask,fscal);
2032 /* Update vectorial force */
2033 fix2 = _mm_macc_ps(dx21,fscal,fix2);
2034 fiy2 = _mm_macc_ps(dy21,fscal,fiy2);
2035 fiz2 = _mm_macc_ps(dz21,fscal,fiz2);
2037 fjx1 = _mm_macc_ps(dx21,fscal,fjx1);
2038 fjy1 = _mm_macc_ps(dy21,fscal,fjy1);
2039 fjz1 = _mm_macc_ps(dz21,fscal,fjz1);
2041 /**************************
2042 * CALCULATE INTERACTIONS *
2043 **************************/
2045 r22 = _mm_mul_ps(rsq22,rinv22);
2046 r22 = _mm_andnot_ps(dummy_mask,r22);
2048 /* Calculate table index by multiplying r with table scale and truncate to integer */
2049 rt = _mm_mul_ps(r22,vftabscale);
2050 vfitab = _mm_cvttps_epi32(rt);
2052 vfeps = _mm_frcz_ps(rt);
2054 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
2056 twovfeps = _mm_add_ps(vfeps,vfeps);
2057 vfitab = _mm_slli_epi32(vfitab,2);
2059 /* CUBIC SPLINE TABLE ELECTROSTATICS */
2060 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
2061 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
2062 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
2063 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
2064 _MM_TRANSPOSE4_PS(Y,F,G,H);
2065 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
2066 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
2067 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq22,FF),_mm_mul_ps(vftabscale,rinv22)));
2071 fscal = _mm_andnot_ps(dummy_mask,fscal);
2073 /* Update vectorial force */
2074 fix2 = _mm_macc_ps(dx22,fscal,fix2);
2075 fiy2 = _mm_macc_ps(dy22,fscal,fiy2);
2076 fiz2 = _mm_macc_ps(dz22,fscal,fiz2);
2078 fjx2 = _mm_macc_ps(dx22,fscal,fjx2);
2079 fjy2 = _mm_macc_ps(dy22,fscal,fjy2);
2080 fjz2 = _mm_macc_ps(dz22,fscal,fjz2);
2082 /**************************
2083 * CALCULATE INTERACTIONS *
2084 **************************/
2086 r23 = _mm_mul_ps(rsq23,rinv23);
2087 r23 = _mm_andnot_ps(dummy_mask,r23);
2089 /* Calculate table index by multiplying r with table scale and truncate to integer */
2090 rt = _mm_mul_ps(r23,vftabscale);
2091 vfitab = _mm_cvttps_epi32(rt);
2093 vfeps = _mm_frcz_ps(rt);
2095 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
2097 twovfeps = _mm_add_ps(vfeps,vfeps);
2098 vfitab = _mm_slli_epi32(vfitab,2);
2100 /* CUBIC SPLINE TABLE ELECTROSTATICS */
2101 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
2102 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
2103 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
2104 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
2105 _MM_TRANSPOSE4_PS(Y,F,G,H);
2106 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
2107 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
2108 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq23,FF),_mm_mul_ps(vftabscale,rinv23)));
2112 fscal = _mm_andnot_ps(dummy_mask,fscal);
2114 /* Update vectorial force */
2115 fix2 = _mm_macc_ps(dx23,fscal,fix2);
2116 fiy2 = _mm_macc_ps(dy23,fscal,fiy2);
2117 fiz2 = _mm_macc_ps(dz23,fscal,fiz2);
2119 fjx3 = _mm_macc_ps(dx23,fscal,fjx3);
2120 fjy3 = _mm_macc_ps(dy23,fscal,fjy3);
2121 fjz3 = _mm_macc_ps(dz23,fscal,fjz3);
2123 /**************************
2124 * CALCULATE INTERACTIONS *
2125 **************************/
2127 r31 = _mm_mul_ps(rsq31,rinv31);
2128 r31 = _mm_andnot_ps(dummy_mask,r31);
2130 /* Calculate table index by multiplying r with table scale and truncate to integer */
2131 rt = _mm_mul_ps(r31,vftabscale);
2132 vfitab = _mm_cvttps_epi32(rt);
2134 vfeps = _mm_frcz_ps(rt);
2136 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
2138 twovfeps = _mm_add_ps(vfeps,vfeps);
2139 vfitab = _mm_slli_epi32(vfitab,2);
2141 /* CUBIC SPLINE TABLE ELECTROSTATICS */
2142 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
2143 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
2144 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
2145 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
2146 _MM_TRANSPOSE4_PS(Y,F,G,H);
2147 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
2148 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
2149 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq31,FF),_mm_mul_ps(vftabscale,rinv31)));
2153 fscal = _mm_andnot_ps(dummy_mask,fscal);
2155 /* Update vectorial force */
2156 fix3 = _mm_macc_ps(dx31,fscal,fix3);
2157 fiy3 = _mm_macc_ps(dy31,fscal,fiy3);
2158 fiz3 = _mm_macc_ps(dz31,fscal,fiz3);
2160 fjx1 = _mm_macc_ps(dx31,fscal,fjx1);
2161 fjy1 = _mm_macc_ps(dy31,fscal,fjy1);
2162 fjz1 = _mm_macc_ps(dz31,fscal,fjz1);
2164 /**************************
2165 * CALCULATE INTERACTIONS *
2166 **************************/
2168 r32 = _mm_mul_ps(rsq32,rinv32);
2169 r32 = _mm_andnot_ps(dummy_mask,r32);
2171 /* Calculate table index by multiplying r with table scale and truncate to integer */
2172 rt = _mm_mul_ps(r32,vftabscale);
2173 vfitab = _mm_cvttps_epi32(rt);
2175 vfeps = _mm_frcz_ps(rt);
2177 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
2179 twovfeps = _mm_add_ps(vfeps,vfeps);
2180 vfitab = _mm_slli_epi32(vfitab,2);
2182 /* CUBIC SPLINE TABLE ELECTROSTATICS */
2183 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
2184 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
2185 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
2186 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
2187 _MM_TRANSPOSE4_PS(Y,F,G,H);
2188 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
2189 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
2190 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq32,FF),_mm_mul_ps(vftabscale,rinv32)));
2194 fscal = _mm_andnot_ps(dummy_mask,fscal);
2196 /* Update vectorial force */
2197 fix3 = _mm_macc_ps(dx32,fscal,fix3);
2198 fiy3 = _mm_macc_ps(dy32,fscal,fiy3);
2199 fiz3 = _mm_macc_ps(dz32,fscal,fiz3);
2201 fjx2 = _mm_macc_ps(dx32,fscal,fjx2);
2202 fjy2 = _mm_macc_ps(dy32,fscal,fjy2);
2203 fjz2 = _mm_macc_ps(dz32,fscal,fjz2);
2205 /**************************
2206 * CALCULATE INTERACTIONS *
2207 **************************/
2209 r33 = _mm_mul_ps(rsq33,rinv33);
2210 r33 = _mm_andnot_ps(dummy_mask,r33);
2212 /* Calculate table index by multiplying r with table scale and truncate to integer */
2213 rt = _mm_mul_ps(r33,vftabscale);
2214 vfitab = _mm_cvttps_epi32(rt);
2216 vfeps = _mm_frcz_ps(rt);
2218 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
2220 twovfeps = _mm_add_ps(vfeps,vfeps);
2221 vfitab = _mm_slli_epi32(vfitab,2);
2223 /* CUBIC SPLINE TABLE ELECTROSTATICS */
2224 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
2225 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
2226 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
2227 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
2228 _MM_TRANSPOSE4_PS(Y,F,G,H);
2229 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
2230 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
2231 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq33,FF),_mm_mul_ps(vftabscale,rinv33)));
2235 fscal = _mm_andnot_ps(dummy_mask,fscal);
2237 /* Update vectorial force */
2238 fix3 = _mm_macc_ps(dx33,fscal,fix3);
2239 fiy3 = _mm_macc_ps(dy33,fscal,fiy3);
2240 fiz3 = _mm_macc_ps(dz33,fscal,fiz3);
2242 fjx3 = _mm_macc_ps(dx33,fscal,fjx3);
2243 fjy3 = _mm_macc_ps(dy33,fscal,fjy3);
2244 fjz3 = _mm_macc_ps(dz33,fscal,fjz3);
2246 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2247 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2248 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2249 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2251 gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
2252 fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2254 /* Inner loop uses 387 flops */
2257 /* End of innermost loop */
2259 gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2260 f+i_coord_offset+DIM,fshift+i_shift_offset);
2262 /* Increment number of inner iterations */
2263 inneriter += j_index_end - j_index_start;
2265 /* Outer loop uses 18 flops */
2268 /* Increment number of outer iterations */
2271 /* Update outer/inner flops */
2273 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_F,outeriter*18 + inneriter*387);