2 * Note: this file was generated by the Gromacs sse2_single kernel generator.
4 * This source code is part of
8 * Copyright (c) 2001-2012, The GROMACS Development Team
10 * Gromacs is a library for molecular simulation and trajectory analysis,
11 * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
12 * a full list of developers and information, check out http://www.gromacs.org
14 * This program is free software; you can redistribute it and/or modify it under
15 * the terms of the GNU Lesser General Public License as published by the Free
16 * Software Foundation; either version 2 of the License, or (at your option) any
19 * To help fund GROMACS development, we humbly ask that you cite
20 * the papers people have written on it - you can find them on the website.
28 #include "../nb_kernel.h"
29 #include "types/simple.h"
33 #include "gmx_math_x86_sse2_single.h"
34 #include "kernelutil_x86_sse2_single.h"
37 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwNone_GeomW4W4_VF_sse2_single
38 * Electrostatics interaction: CubicSplineTable
39 * VdW interaction: None
40 * Geometry: Water4-Water4
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecCSTab_VdwNone_GeomW4W4_VF_sse2_single
45 (t_nblist * gmx_restrict nlist,
46 rvec * gmx_restrict xx,
47 rvec * gmx_restrict ff,
48 t_forcerec * gmx_restrict fr,
49 t_mdatoms * gmx_restrict mdatoms,
50 nb_kernel_data_t * gmx_restrict kernel_data,
51 t_nrnb * gmx_restrict nrnb)
53 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
54 * just 0 for non-waters.
55 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
56 * jnr indices corresponding to data put in the four positions in the SIMD register.
58 int i_shift_offset,i_coord_offset,outeriter,inneriter;
59 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
60 int jnrA,jnrB,jnrC,jnrD;
61 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
62 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
63 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
65 real *shiftvec,*fshift,*x,*f;
66 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
68 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
70 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
72 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
74 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
75 int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
76 __m128 jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
77 int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
78 __m128 jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
79 int vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
80 __m128 jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
81 __m128 dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
82 __m128 dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
83 __m128 dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
84 __m128 dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
85 __m128 dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
86 __m128 dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
87 __m128 dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
88 __m128 dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
89 __m128 dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
90 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
93 __m128i ifour = _mm_set1_epi32(4);
94 __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
96 __m128 dummy_mask,cutoff_mask;
97 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
98 __m128 one = _mm_set1_ps(1.0);
99 __m128 two = _mm_set1_ps(2.0);
105 jindex = nlist->jindex;
107 shiftidx = nlist->shift;
109 shiftvec = fr->shift_vec[0];
110 fshift = fr->fshift[0];
111 facel = _mm_set1_ps(fr->epsfac);
112 charge = mdatoms->chargeA;
114 vftab = kernel_data->table_elec->data;
115 vftabscale = _mm_set1_ps(kernel_data->table_elec->scale);
117 /* Setup water-specific parameters */
118 inr = nlist->iinr[0];
119 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
120 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
121 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
123 jq1 = _mm_set1_ps(charge[inr+1]);
124 jq2 = _mm_set1_ps(charge[inr+2]);
125 jq3 = _mm_set1_ps(charge[inr+3]);
126 qq11 = _mm_mul_ps(iq1,jq1);
127 qq12 = _mm_mul_ps(iq1,jq2);
128 qq13 = _mm_mul_ps(iq1,jq3);
129 qq21 = _mm_mul_ps(iq2,jq1);
130 qq22 = _mm_mul_ps(iq2,jq2);
131 qq23 = _mm_mul_ps(iq2,jq3);
132 qq31 = _mm_mul_ps(iq3,jq1);
133 qq32 = _mm_mul_ps(iq3,jq2);
134 qq33 = _mm_mul_ps(iq3,jq3);
136 /* Avoid stupid compiler warnings */
137 jnrA = jnrB = jnrC = jnrD = 0;
146 for(iidx=0;iidx<4*DIM;iidx++)
151 /* Start outer loop over neighborlists */
152 for(iidx=0; iidx<nri; iidx++)
154 /* Load shift vector for this list */
155 i_shift_offset = DIM*shiftidx[iidx];
157 /* Load limits for loop over neighbors */
158 j_index_start = jindex[iidx];
159 j_index_end = jindex[iidx+1];
161 /* Get outer coordinate index */
163 i_coord_offset = DIM*inr;
165 /* Load i particle coords and add shift vector */
166 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
167 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
169 fix1 = _mm_setzero_ps();
170 fiy1 = _mm_setzero_ps();
171 fiz1 = _mm_setzero_ps();
172 fix2 = _mm_setzero_ps();
173 fiy2 = _mm_setzero_ps();
174 fiz2 = _mm_setzero_ps();
175 fix3 = _mm_setzero_ps();
176 fiy3 = _mm_setzero_ps();
177 fiz3 = _mm_setzero_ps();
179 /* Reset potential sums */
180 velecsum = _mm_setzero_ps();
182 /* Start inner kernel loop */
183 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
186 /* Get j neighbor index, and coordinate index */
191 j_coord_offsetA = DIM*jnrA;
192 j_coord_offsetB = DIM*jnrB;
193 j_coord_offsetC = DIM*jnrC;
194 j_coord_offsetD = DIM*jnrD;
196 /* load j atom coordinates */
197 gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
198 x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
199 &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
201 /* Calculate displacement vector */
202 dx11 = _mm_sub_ps(ix1,jx1);
203 dy11 = _mm_sub_ps(iy1,jy1);
204 dz11 = _mm_sub_ps(iz1,jz1);
205 dx12 = _mm_sub_ps(ix1,jx2);
206 dy12 = _mm_sub_ps(iy1,jy2);
207 dz12 = _mm_sub_ps(iz1,jz2);
208 dx13 = _mm_sub_ps(ix1,jx3);
209 dy13 = _mm_sub_ps(iy1,jy3);
210 dz13 = _mm_sub_ps(iz1,jz3);
211 dx21 = _mm_sub_ps(ix2,jx1);
212 dy21 = _mm_sub_ps(iy2,jy1);
213 dz21 = _mm_sub_ps(iz2,jz1);
214 dx22 = _mm_sub_ps(ix2,jx2);
215 dy22 = _mm_sub_ps(iy2,jy2);
216 dz22 = _mm_sub_ps(iz2,jz2);
217 dx23 = _mm_sub_ps(ix2,jx3);
218 dy23 = _mm_sub_ps(iy2,jy3);
219 dz23 = _mm_sub_ps(iz2,jz3);
220 dx31 = _mm_sub_ps(ix3,jx1);
221 dy31 = _mm_sub_ps(iy3,jy1);
222 dz31 = _mm_sub_ps(iz3,jz1);
223 dx32 = _mm_sub_ps(ix3,jx2);
224 dy32 = _mm_sub_ps(iy3,jy2);
225 dz32 = _mm_sub_ps(iz3,jz2);
226 dx33 = _mm_sub_ps(ix3,jx3);
227 dy33 = _mm_sub_ps(iy3,jy3);
228 dz33 = _mm_sub_ps(iz3,jz3);
230 /* Calculate squared distance and things based on it */
231 rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
232 rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
233 rsq13 = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
234 rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
235 rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
236 rsq23 = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
237 rsq31 = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
238 rsq32 = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
239 rsq33 = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
241 rinv11 = gmx_mm_invsqrt_ps(rsq11);
242 rinv12 = gmx_mm_invsqrt_ps(rsq12);
243 rinv13 = gmx_mm_invsqrt_ps(rsq13);
244 rinv21 = gmx_mm_invsqrt_ps(rsq21);
245 rinv22 = gmx_mm_invsqrt_ps(rsq22);
246 rinv23 = gmx_mm_invsqrt_ps(rsq23);
247 rinv31 = gmx_mm_invsqrt_ps(rsq31);
248 rinv32 = gmx_mm_invsqrt_ps(rsq32);
249 rinv33 = gmx_mm_invsqrt_ps(rsq33);
251 fjx1 = _mm_setzero_ps();
252 fjy1 = _mm_setzero_ps();
253 fjz1 = _mm_setzero_ps();
254 fjx2 = _mm_setzero_ps();
255 fjy2 = _mm_setzero_ps();
256 fjz2 = _mm_setzero_ps();
257 fjx3 = _mm_setzero_ps();
258 fjy3 = _mm_setzero_ps();
259 fjz3 = _mm_setzero_ps();
261 /**************************
262 * CALCULATE INTERACTIONS *
263 **************************/
265 r11 = _mm_mul_ps(rsq11,rinv11);
267 /* Calculate table index by multiplying r with table scale and truncate to integer */
268 rt = _mm_mul_ps(r11,vftabscale);
269 vfitab = _mm_cvttps_epi32(rt);
270 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
271 vfitab = _mm_slli_epi32(vfitab,2);
273 /* CUBIC SPLINE TABLE ELECTROSTATICS */
274 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
275 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
276 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
277 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
278 _MM_TRANSPOSE4_PS(Y,F,G,H);
279 Heps = _mm_mul_ps(vfeps,H);
280 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
281 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
282 velec = _mm_mul_ps(qq11,VV);
283 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
284 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq11,FF),_mm_mul_ps(vftabscale,rinv11)));
286 /* Update potential sum for this i atom from the interaction with this j atom. */
287 velecsum = _mm_add_ps(velecsum,velec);
291 /* Calculate temporary vectorial force */
292 tx = _mm_mul_ps(fscal,dx11);
293 ty = _mm_mul_ps(fscal,dy11);
294 tz = _mm_mul_ps(fscal,dz11);
296 /* Update vectorial force */
297 fix1 = _mm_add_ps(fix1,tx);
298 fiy1 = _mm_add_ps(fiy1,ty);
299 fiz1 = _mm_add_ps(fiz1,tz);
301 fjx1 = _mm_add_ps(fjx1,tx);
302 fjy1 = _mm_add_ps(fjy1,ty);
303 fjz1 = _mm_add_ps(fjz1,tz);
305 /**************************
306 * CALCULATE INTERACTIONS *
307 **************************/
309 r12 = _mm_mul_ps(rsq12,rinv12);
311 /* Calculate table index by multiplying r with table scale and truncate to integer */
312 rt = _mm_mul_ps(r12,vftabscale);
313 vfitab = _mm_cvttps_epi32(rt);
314 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
315 vfitab = _mm_slli_epi32(vfitab,2);
317 /* CUBIC SPLINE TABLE ELECTROSTATICS */
318 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
319 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
320 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
321 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
322 _MM_TRANSPOSE4_PS(Y,F,G,H);
323 Heps = _mm_mul_ps(vfeps,H);
324 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
325 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
326 velec = _mm_mul_ps(qq12,VV);
327 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
328 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq12,FF),_mm_mul_ps(vftabscale,rinv12)));
330 /* Update potential sum for this i atom from the interaction with this j atom. */
331 velecsum = _mm_add_ps(velecsum,velec);
335 /* Calculate temporary vectorial force */
336 tx = _mm_mul_ps(fscal,dx12);
337 ty = _mm_mul_ps(fscal,dy12);
338 tz = _mm_mul_ps(fscal,dz12);
340 /* Update vectorial force */
341 fix1 = _mm_add_ps(fix1,tx);
342 fiy1 = _mm_add_ps(fiy1,ty);
343 fiz1 = _mm_add_ps(fiz1,tz);
345 fjx2 = _mm_add_ps(fjx2,tx);
346 fjy2 = _mm_add_ps(fjy2,ty);
347 fjz2 = _mm_add_ps(fjz2,tz);
349 /**************************
350 * CALCULATE INTERACTIONS *
351 **************************/
353 r13 = _mm_mul_ps(rsq13,rinv13);
355 /* Calculate table index by multiplying r with table scale and truncate to integer */
356 rt = _mm_mul_ps(r13,vftabscale);
357 vfitab = _mm_cvttps_epi32(rt);
358 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
359 vfitab = _mm_slli_epi32(vfitab,2);
361 /* CUBIC SPLINE TABLE ELECTROSTATICS */
362 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
363 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
364 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
365 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
366 _MM_TRANSPOSE4_PS(Y,F,G,H);
367 Heps = _mm_mul_ps(vfeps,H);
368 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
369 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
370 velec = _mm_mul_ps(qq13,VV);
371 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
372 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq13,FF),_mm_mul_ps(vftabscale,rinv13)));
374 /* Update potential sum for this i atom from the interaction with this j atom. */
375 velecsum = _mm_add_ps(velecsum,velec);
379 /* Calculate temporary vectorial force */
380 tx = _mm_mul_ps(fscal,dx13);
381 ty = _mm_mul_ps(fscal,dy13);
382 tz = _mm_mul_ps(fscal,dz13);
384 /* Update vectorial force */
385 fix1 = _mm_add_ps(fix1,tx);
386 fiy1 = _mm_add_ps(fiy1,ty);
387 fiz1 = _mm_add_ps(fiz1,tz);
389 fjx3 = _mm_add_ps(fjx3,tx);
390 fjy3 = _mm_add_ps(fjy3,ty);
391 fjz3 = _mm_add_ps(fjz3,tz);
393 /**************************
394 * CALCULATE INTERACTIONS *
395 **************************/
397 r21 = _mm_mul_ps(rsq21,rinv21);
399 /* Calculate table index by multiplying r with table scale and truncate to integer */
400 rt = _mm_mul_ps(r21,vftabscale);
401 vfitab = _mm_cvttps_epi32(rt);
402 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
403 vfitab = _mm_slli_epi32(vfitab,2);
405 /* CUBIC SPLINE TABLE ELECTROSTATICS */
406 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
407 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
408 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
409 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
410 _MM_TRANSPOSE4_PS(Y,F,G,H);
411 Heps = _mm_mul_ps(vfeps,H);
412 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
413 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
414 velec = _mm_mul_ps(qq21,VV);
415 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
416 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq21,FF),_mm_mul_ps(vftabscale,rinv21)));
418 /* Update potential sum for this i atom from the interaction with this j atom. */
419 velecsum = _mm_add_ps(velecsum,velec);
423 /* Calculate temporary vectorial force */
424 tx = _mm_mul_ps(fscal,dx21);
425 ty = _mm_mul_ps(fscal,dy21);
426 tz = _mm_mul_ps(fscal,dz21);
428 /* Update vectorial force */
429 fix2 = _mm_add_ps(fix2,tx);
430 fiy2 = _mm_add_ps(fiy2,ty);
431 fiz2 = _mm_add_ps(fiz2,tz);
433 fjx1 = _mm_add_ps(fjx1,tx);
434 fjy1 = _mm_add_ps(fjy1,ty);
435 fjz1 = _mm_add_ps(fjz1,tz);
437 /**************************
438 * CALCULATE INTERACTIONS *
439 **************************/
441 r22 = _mm_mul_ps(rsq22,rinv22);
443 /* Calculate table index by multiplying r with table scale and truncate to integer */
444 rt = _mm_mul_ps(r22,vftabscale);
445 vfitab = _mm_cvttps_epi32(rt);
446 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
447 vfitab = _mm_slli_epi32(vfitab,2);
449 /* CUBIC SPLINE TABLE ELECTROSTATICS */
450 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
451 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
452 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
453 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
454 _MM_TRANSPOSE4_PS(Y,F,G,H);
455 Heps = _mm_mul_ps(vfeps,H);
456 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
457 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
458 velec = _mm_mul_ps(qq22,VV);
459 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
460 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq22,FF),_mm_mul_ps(vftabscale,rinv22)));
462 /* Update potential sum for this i atom from the interaction with this j atom. */
463 velecsum = _mm_add_ps(velecsum,velec);
467 /* Calculate temporary vectorial force */
468 tx = _mm_mul_ps(fscal,dx22);
469 ty = _mm_mul_ps(fscal,dy22);
470 tz = _mm_mul_ps(fscal,dz22);
472 /* Update vectorial force */
473 fix2 = _mm_add_ps(fix2,tx);
474 fiy2 = _mm_add_ps(fiy2,ty);
475 fiz2 = _mm_add_ps(fiz2,tz);
477 fjx2 = _mm_add_ps(fjx2,tx);
478 fjy2 = _mm_add_ps(fjy2,ty);
479 fjz2 = _mm_add_ps(fjz2,tz);
481 /**************************
482 * CALCULATE INTERACTIONS *
483 **************************/
485 r23 = _mm_mul_ps(rsq23,rinv23);
487 /* Calculate table index by multiplying r with table scale and truncate to integer */
488 rt = _mm_mul_ps(r23,vftabscale);
489 vfitab = _mm_cvttps_epi32(rt);
490 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
491 vfitab = _mm_slli_epi32(vfitab,2);
493 /* CUBIC SPLINE TABLE ELECTROSTATICS */
494 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
495 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
496 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
497 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
498 _MM_TRANSPOSE4_PS(Y,F,G,H);
499 Heps = _mm_mul_ps(vfeps,H);
500 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
501 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
502 velec = _mm_mul_ps(qq23,VV);
503 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
504 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq23,FF),_mm_mul_ps(vftabscale,rinv23)));
506 /* Update potential sum for this i atom from the interaction with this j atom. */
507 velecsum = _mm_add_ps(velecsum,velec);
511 /* Calculate temporary vectorial force */
512 tx = _mm_mul_ps(fscal,dx23);
513 ty = _mm_mul_ps(fscal,dy23);
514 tz = _mm_mul_ps(fscal,dz23);
516 /* Update vectorial force */
517 fix2 = _mm_add_ps(fix2,tx);
518 fiy2 = _mm_add_ps(fiy2,ty);
519 fiz2 = _mm_add_ps(fiz2,tz);
521 fjx3 = _mm_add_ps(fjx3,tx);
522 fjy3 = _mm_add_ps(fjy3,ty);
523 fjz3 = _mm_add_ps(fjz3,tz);
525 /**************************
526 * CALCULATE INTERACTIONS *
527 **************************/
529 r31 = _mm_mul_ps(rsq31,rinv31);
531 /* Calculate table index by multiplying r with table scale and truncate to integer */
532 rt = _mm_mul_ps(r31,vftabscale);
533 vfitab = _mm_cvttps_epi32(rt);
534 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
535 vfitab = _mm_slli_epi32(vfitab,2);
537 /* CUBIC SPLINE TABLE ELECTROSTATICS */
538 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
539 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
540 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
541 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
542 _MM_TRANSPOSE4_PS(Y,F,G,H);
543 Heps = _mm_mul_ps(vfeps,H);
544 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
545 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
546 velec = _mm_mul_ps(qq31,VV);
547 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
548 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq31,FF),_mm_mul_ps(vftabscale,rinv31)));
550 /* Update potential sum for this i atom from the interaction with this j atom. */
551 velecsum = _mm_add_ps(velecsum,velec);
555 /* Calculate temporary vectorial force */
556 tx = _mm_mul_ps(fscal,dx31);
557 ty = _mm_mul_ps(fscal,dy31);
558 tz = _mm_mul_ps(fscal,dz31);
560 /* Update vectorial force */
561 fix3 = _mm_add_ps(fix3,tx);
562 fiy3 = _mm_add_ps(fiy3,ty);
563 fiz3 = _mm_add_ps(fiz3,tz);
565 fjx1 = _mm_add_ps(fjx1,tx);
566 fjy1 = _mm_add_ps(fjy1,ty);
567 fjz1 = _mm_add_ps(fjz1,tz);
569 /**************************
570 * CALCULATE INTERACTIONS *
571 **************************/
573 r32 = _mm_mul_ps(rsq32,rinv32);
575 /* Calculate table index by multiplying r with table scale and truncate to integer */
576 rt = _mm_mul_ps(r32,vftabscale);
577 vfitab = _mm_cvttps_epi32(rt);
578 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
579 vfitab = _mm_slli_epi32(vfitab,2);
581 /* CUBIC SPLINE TABLE ELECTROSTATICS */
582 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
583 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
584 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
585 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
586 _MM_TRANSPOSE4_PS(Y,F,G,H);
587 Heps = _mm_mul_ps(vfeps,H);
588 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
589 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
590 velec = _mm_mul_ps(qq32,VV);
591 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
592 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq32,FF),_mm_mul_ps(vftabscale,rinv32)));
594 /* Update potential sum for this i atom from the interaction with this j atom. */
595 velecsum = _mm_add_ps(velecsum,velec);
599 /* Calculate temporary vectorial force */
600 tx = _mm_mul_ps(fscal,dx32);
601 ty = _mm_mul_ps(fscal,dy32);
602 tz = _mm_mul_ps(fscal,dz32);
604 /* Update vectorial force */
605 fix3 = _mm_add_ps(fix3,tx);
606 fiy3 = _mm_add_ps(fiy3,ty);
607 fiz3 = _mm_add_ps(fiz3,tz);
609 fjx2 = _mm_add_ps(fjx2,tx);
610 fjy2 = _mm_add_ps(fjy2,ty);
611 fjz2 = _mm_add_ps(fjz2,tz);
613 /**************************
614 * CALCULATE INTERACTIONS *
615 **************************/
617 r33 = _mm_mul_ps(rsq33,rinv33);
619 /* Calculate table index by multiplying r with table scale and truncate to integer */
620 rt = _mm_mul_ps(r33,vftabscale);
621 vfitab = _mm_cvttps_epi32(rt);
622 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
623 vfitab = _mm_slli_epi32(vfitab,2);
625 /* CUBIC SPLINE TABLE ELECTROSTATICS */
626 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
627 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
628 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
629 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
630 _MM_TRANSPOSE4_PS(Y,F,G,H);
631 Heps = _mm_mul_ps(vfeps,H);
632 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
633 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
634 velec = _mm_mul_ps(qq33,VV);
635 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
636 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq33,FF),_mm_mul_ps(vftabscale,rinv33)));
638 /* Update potential sum for this i atom from the interaction with this j atom. */
639 velecsum = _mm_add_ps(velecsum,velec);
643 /* Calculate temporary vectorial force */
644 tx = _mm_mul_ps(fscal,dx33);
645 ty = _mm_mul_ps(fscal,dy33);
646 tz = _mm_mul_ps(fscal,dz33);
648 /* Update vectorial force */
649 fix3 = _mm_add_ps(fix3,tx);
650 fiy3 = _mm_add_ps(fiy3,ty);
651 fiz3 = _mm_add_ps(fiz3,tz);
653 fjx3 = _mm_add_ps(fjx3,tx);
654 fjy3 = _mm_add_ps(fjy3,ty);
655 fjz3 = _mm_add_ps(fjz3,tz);
657 fjptrA = f+j_coord_offsetA;
658 fjptrB = f+j_coord_offsetB;
659 fjptrC = f+j_coord_offsetC;
660 fjptrD = f+j_coord_offsetD;
662 gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
663 fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
665 /* Inner loop uses 387 flops */
671 /* Get j neighbor index, and coordinate index */
672 jnrlistA = jjnr[jidx];
673 jnrlistB = jjnr[jidx+1];
674 jnrlistC = jjnr[jidx+2];
675 jnrlistD = jjnr[jidx+3];
676 /* Sign of each element will be negative for non-real atoms.
677 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
678 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
680 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
681 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
682 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
683 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
684 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
685 j_coord_offsetA = DIM*jnrA;
686 j_coord_offsetB = DIM*jnrB;
687 j_coord_offsetC = DIM*jnrC;
688 j_coord_offsetD = DIM*jnrD;
690 /* load j atom coordinates */
691 gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
692 x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
693 &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
695 /* Calculate displacement vector */
696 dx11 = _mm_sub_ps(ix1,jx1);
697 dy11 = _mm_sub_ps(iy1,jy1);
698 dz11 = _mm_sub_ps(iz1,jz1);
699 dx12 = _mm_sub_ps(ix1,jx2);
700 dy12 = _mm_sub_ps(iy1,jy2);
701 dz12 = _mm_sub_ps(iz1,jz2);
702 dx13 = _mm_sub_ps(ix1,jx3);
703 dy13 = _mm_sub_ps(iy1,jy3);
704 dz13 = _mm_sub_ps(iz1,jz3);
705 dx21 = _mm_sub_ps(ix2,jx1);
706 dy21 = _mm_sub_ps(iy2,jy1);
707 dz21 = _mm_sub_ps(iz2,jz1);
708 dx22 = _mm_sub_ps(ix2,jx2);
709 dy22 = _mm_sub_ps(iy2,jy2);
710 dz22 = _mm_sub_ps(iz2,jz2);
711 dx23 = _mm_sub_ps(ix2,jx3);
712 dy23 = _mm_sub_ps(iy2,jy3);
713 dz23 = _mm_sub_ps(iz2,jz3);
714 dx31 = _mm_sub_ps(ix3,jx1);
715 dy31 = _mm_sub_ps(iy3,jy1);
716 dz31 = _mm_sub_ps(iz3,jz1);
717 dx32 = _mm_sub_ps(ix3,jx2);
718 dy32 = _mm_sub_ps(iy3,jy2);
719 dz32 = _mm_sub_ps(iz3,jz2);
720 dx33 = _mm_sub_ps(ix3,jx3);
721 dy33 = _mm_sub_ps(iy3,jy3);
722 dz33 = _mm_sub_ps(iz3,jz3);
724 /* Calculate squared distance and things based on it */
725 rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
726 rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
727 rsq13 = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
728 rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
729 rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
730 rsq23 = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
731 rsq31 = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
732 rsq32 = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
733 rsq33 = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
735 rinv11 = gmx_mm_invsqrt_ps(rsq11);
736 rinv12 = gmx_mm_invsqrt_ps(rsq12);
737 rinv13 = gmx_mm_invsqrt_ps(rsq13);
738 rinv21 = gmx_mm_invsqrt_ps(rsq21);
739 rinv22 = gmx_mm_invsqrt_ps(rsq22);
740 rinv23 = gmx_mm_invsqrt_ps(rsq23);
741 rinv31 = gmx_mm_invsqrt_ps(rsq31);
742 rinv32 = gmx_mm_invsqrt_ps(rsq32);
743 rinv33 = gmx_mm_invsqrt_ps(rsq33);
745 fjx1 = _mm_setzero_ps();
746 fjy1 = _mm_setzero_ps();
747 fjz1 = _mm_setzero_ps();
748 fjx2 = _mm_setzero_ps();
749 fjy2 = _mm_setzero_ps();
750 fjz2 = _mm_setzero_ps();
751 fjx3 = _mm_setzero_ps();
752 fjy3 = _mm_setzero_ps();
753 fjz3 = _mm_setzero_ps();
755 /**************************
756 * CALCULATE INTERACTIONS *
757 **************************/
759 r11 = _mm_mul_ps(rsq11,rinv11);
760 r11 = _mm_andnot_ps(dummy_mask,r11);
762 /* Calculate table index by multiplying r with table scale and truncate to integer */
763 rt = _mm_mul_ps(r11,vftabscale);
764 vfitab = _mm_cvttps_epi32(rt);
765 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
766 vfitab = _mm_slli_epi32(vfitab,2);
768 /* CUBIC SPLINE TABLE ELECTROSTATICS */
769 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
770 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
771 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
772 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
773 _MM_TRANSPOSE4_PS(Y,F,G,H);
774 Heps = _mm_mul_ps(vfeps,H);
775 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
776 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
777 velec = _mm_mul_ps(qq11,VV);
778 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
779 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq11,FF),_mm_mul_ps(vftabscale,rinv11)));
781 /* Update potential sum for this i atom from the interaction with this j atom. */
782 velec = _mm_andnot_ps(dummy_mask,velec);
783 velecsum = _mm_add_ps(velecsum,velec);
787 fscal = _mm_andnot_ps(dummy_mask,fscal);
789 /* Calculate temporary vectorial force */
790 tx = _mm_mul_ps(fscal,dx11);
791 ty = _mm_mul_ps(fscal,dy11);
792 tz = _mm_mul_ps(fscal,dz11);
794 /* Update vectorial force */
795 fix1 = _mm_add_ps(fix1,tx);
796 fiy1 = _mm_add_ps(fiy1,ty);
797 fiz1 = _mm_add_ps(fiz1,tz);
799 fjx1 = _mm_add_ps(fjx1,tx);
800 fjy1 = _mm_add_ps(fjy1,ty);
801 fjz1 = _mm_add_ps(fjz1,tz);
803 /**************************
804 * CALCULATE INTERACTIONS *
805 **************************/
807 r12 = _mm_mul_ps(rsq12,rinv12);
808 r12 = _mm_andnot_ps(dummy_mask,r12);
810 /* Calculate table index by multiplying r with table scale and truncate to integer */
811 rt = _mm_mul_ps(r12,vftabscale);
812 vfitab = _mm_cvttps_epi32(rt);
813 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
814 vfitab = _mm_slli_epi32(vfitab,2);
816 /* CUBIC SPLINE TABLE ELECTROSTATICS */
817 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
818 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
819 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
820 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
821 _MM_TRANSPOSE4_PS(Y,F,G,H);
822 Heps = _mm_mul_ps(vfeps,H);
823 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
824 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
825 velec = _mm_mul_ps(qq12,VV);
826 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
827 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq12,FF),_mm_mul_ps(vftabscale,rinv12)));
829 /* Update potential sum for this i atom from the interaction with this j atom. */
830 velec = _mm_andnot_ps(dummy_mask,velec);
831 velecsum = _mm_add_ps(velecsum,velec);
835 fscal = _mm_andnot_ps(dummy_mask,fscal);
837 /* Calculate temporary vectorial force */
838 tx = _mm_mul_ps(fscal,dx12);
839 ty = _mm_mul_ps(fscal,dy12);
840 tz = _mm_mul_ps(fscal,dz12);
842 /* Update vectorial force */
843 fix1 = _mm_add_ps(fix1,tx);
844 fiy1 = _mm_add_ps(fiy1,ty);
845 fiz1 = _mm_add_ps(fiz1,tz);
847 fjx2 = _mm_add_ps(fjx2,tx);
848 fjy2 = _mm_add_ps(fjy2,ty);
849 fjz2 = _mm_add_ps(fjz2,tz);
851 /**************************
852 * CALCULATE INTERACTIONS *
853 **************************/
855 r13 = _mm_mul_ps(rsq13,rinv13);
856 r13 = _mm_andnot_ps(dummy_mask,r13);
858 /* Calculate table index by multiplying r with table scale and truncate to integer */
859 rt = _mm_mul_ps(r13,vftabscale);
860 vfitab = _mm_cvttps_epi32(rt);
861 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
862 vfitab = _mm_slli_epi32(vfitab,2);
864 /* CUBIC SPLINE TABLE ELECTROSTATICS */
865 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
866 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
867 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
868 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
869 _MM_TRANSPOSE4_PS(Y,F,G,H);
870 Heps = _mm_mul_ps(vfeps,H);
871 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
872 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
873 velec = _mm_mul_ps(qq13,VV);
874 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
875 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq13,FF),_mm_mul_ps(vftabscale,rinv13)));
877 /* Update potential sum for this i atom from the interaction with this j atom. */
878 velec = _mm_andnot_ps(dummy_mask,velec);
879 velecsum = _mm_add_ps(velecsum,velec);
883 fscal = _mm_andnot_ps(dummy_mask,fscal);
885 /* Calculate temporary vectorial force */
886 tx = _mm_mul_ps(fscal,dx13);
887 ty = _mm_mul_ps(fscal,dy13);
888 tz = _mm_mul_ps(fscal,dz13);
890 /* Update vectorial force */
891 fix1 = _mm_add_ps(fix1,tx);
892 fiy1 = _mm_add_ps(fiy1,ty);
893 fiz1 = _mm_add_ps(fiz1,tz);
895 fjx3 = _mm_add_ps(fjx3,tx);
896 fjy3 = _mm_add_ps(fjy3,ty);
897 fjz3 = _mm_add_ps(fjz3,tz);
899 /**************************
900 * CALCULATE INTERACTIONS *
901 **************************/
903 r21 = _mm_mul_ps(rsq21,rinv21);
904 r21 = _mm_andnot_ps(dummy_mask,r21);
906 /* Calculate table index by multiplying r with table scale and truncate to integer */
907 rt = _mm_mul_ps(r21,vftabscale);
908 vfitab = _mm_cvttps_epi32(rt);
909 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
910 vfitab = _mm_slli_epi32(vfitab,2);
912 /* CUBIC SPLINE TABLE ELECTROSTATICS */
913 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
914 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
915 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
916 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
917 _MM_TRANSPOSE4_PS(Y,F,G,H);
918 Heps = _mm_mul_ps(vfeps,H);
919 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
920 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
921 velec = _mm_mul_ps(qq21,VV);
922 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
923 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq21,FF),_mm_mul_ps(vftabscale,rinv21)));
925 /* Update potential sum for this i atom from the interaction with this j atom. */
926 velec = _mm_andnot_ps(dummy_mask,velec);
927 velecsum = _mm_add_ps(velecsum,velec);
931 fscal = _mm_andnot_ps(dummy_mask,fscal);
933 /* Calculate temporary vectorial force */
934 tx = _mm_mul_ps(fscal,dx21);
935 ty = _mm_mul_ps(fscal,dy21);
936 tz = _mm_mul_ps(fscal,dz21);
938 /* Update vectorial force */
939 fix2 = _mm_add_ps(fix2,tx);
940 fiy2 = _mm_add_ps(fiy2,ty);
941 fiz2 = _mm_add_ps(fiz2,tz);
943 fjx1 = _mm_add_ps(fjx1,tx);
944 fjy1 = _mm_add_ps(fjy1,ty);
945 fjz1 = _mm_add_ps(fjz1,tz);
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);
957 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
958 vfitab = _mm_slli_epi32(vfitab,2);
960 /* CUBIC SPLINE TABLE ELECTROSTATICS */
961 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
962 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
963 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
964 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
965 _MM_TRANSPOSE4_PS(Y,F,G,H);
966 Heps = _mm_mul_ps(vfeps,H);
967 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
968 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
969 velec = _mm_mul_ps(qq22,VV);
970 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
971 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq22,FF),_mm_mul_ps(vftabscale,rinv22)));
973 /* Update potential sum for this i atom from the interaction with this j atom. */
974 velec = _mm_andnot_ps(dummy_mask,velec);
975 velecsum = _mm_add_ps(velecsum,velec);
979 fscal = _mm_andnot_ps(dummy_mask,fscal);
981 /* Calculate temporary vectorial force */
982 tx = _mm_mul_ps(fscal,dx22);
983 ty = _mm_mul_ps(fscal,dy22);
984 tz = _mm_mul_ps(fscal,dz22);
986 /* Update vectorial force */
987 fix2 = _mm_add_ps(fix2,tx);
988 fiy2 = _mm_add_ps(fiy2,ty);
989 fiz2 = _mm_add_ps(fiz2,tz);
991 fjx2 = _mm_add_ps(fjx2,tx);
992 fjy2 = _mm_add_ps(fjy2,ty);
993 fjz2 = _mm_add_ps(fjz2,tz);
995 /**************************
996 * CALCULATE INTERACTIONS *
997 **************************/
999 r23 = _mm_mul_ps(rsq23,rinv23);
1000 r23 = _mm_andnot_ps(dummy_mask,r23);
1002 /* Calculate table index by multiplying r with table scale and truncate to integer */
1003 rt = _mm_mul_ps(r23,vftabscale);
1004 vfitab = _mm_cvttps_epi32(rt);
1005 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1006 vfitab = _mm_slli_epi32(vfitab,2);
1008 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1009 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1010 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1011 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1012 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1013 _MM_TRANSPOSE4_PS(Y,F,G,H);
1014 Heps = _mm_mul_ps(vfeps,H);
1015 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1016 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
1017 velec = _mm_mul_ps(qq23,VV);
1018 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1019 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq23,FF),_mm_mul_ps(vftabscale,rinv23)));
1021 /* Update potential sum for this i atom from the interaction with this j atom. */
1022 velec = _mm_andnot_ps(dummy_mask,velec);
1023 velecsum = _mm_add_ps(velecsum,velec);
1027 fscal = _mm_andnot_ps(dummy_mask,fscal);
1029 /* Calculate temporary vectorial force */
1030 tx = _mm_mul_ps(fscal,dx23);
1031 ty = _mm_mul_ps(fscal,dy23);
1032 tz = _mm_mul_ps(fscal,dz23);
1034 /* Update vectorial force */
1035 fix2 = _mm_add_ps(fix2,tx);
1036 fiy2 = _mm_add_ps(fiy2,ty);
1037 fiz2 = _mm_add_ps(fiz2,tz);
1039 fjx3 = _mm_add_ps(fjx3,tx);
1040 fjy3 = _mm_add_ps(fjy3,ty);
1041 fjz3 = _mm_add_ps(fjz3,tz);
1043 /**************************
1044 * CALCULATE INTERACTIONS *
1045 **************************/
1047 r31 = _mm_mul_ps(rsq31,rinv31);
1048 r31 = _mm_andnot_ps(dummy_mask,r31);
1050 /* Calculate table index by multiplying r with table scale and truncate to integer */
1051 rt = _mm_mul_ps(r31,vftabscale);
1052 vfitab = _mm_cvttps_epi32(rt);
1053 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1054 vfitab = _mm_slli_epi32(vfitab,2);
1056 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1057 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1058 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1059 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1060 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1061 _MM_TRANSPOSE4_PS(Y,F,G,H);
1062 Heps = _mm_mul_ps(vfeps,H);
1063 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1064 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
1065 velec = _mm_mul_ps(qq31,VV);
1066 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1067 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq31,FF),_mm_mul_ps(vftabscale,rinv31)));
1069 /* Update potential sum for this i atom from the interaction with this j atom. */
1070 velec = _mm_andnot_ps(dummy_mask,velec);
1071 velecsum = _mm_add_ps(velecsum,velec);
1075 fscal = _mm_andnot_ps(dummy_mask,fscal);
1077 /* Calculate temporary vectorial force */
1078 tx = _mm_mul_ps(fscal,dx31);
1079 ty = _mm_mul_ps(fscal,dy31);
1080 tz = _mm_mul_ps(fscal,dz31);
1082 /* Update vectorial force */
1083 fix3 = _mm_add_ps(fix3,tx);
1084 fiy3 = _mm_add_ps(fiy3,ty);
1085 fiz3 = _mm_add_ps(fiz3,tz);
1087 fjx1 = _mm_add_ps(fjx1,tx);
1088 fjy1 = _mm_add_ps(fjy1,ty);
1089 fjz1 = _mm_add_ps(fjz1,tz);
1091 /**************************
1092 * CALCULATE INTERACTIONS *
1093 **************************/
1095 r32 = _mm_mul_ps(rsq32,rinv32);
1096 r32 = _mm_andnot_ps(dummy_mask,r32);
1098 /* Calculate table index by multiplying r with table scale and truncate to integer */
1099 rt = _mm_mul_ps(r32,vftabscale);
1100 vfitab = _mm_cvttps_epi32(rt);
1101 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1102 vfitab = _mm_slli_epi32(vfitab,2);
1104 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1105 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1106 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1107 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1108 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1109 _MM_TRANSPOSE4_PS(Y,F,G,H);
1110 Heps = _mm_mul_ps(vfeps,H);
1111 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1112 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
1113 velec = _mm_mul_ps(qq32,VV);
1114 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1115 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq32,FF),_mm_mul_ps(vftabscale,rinv32)));
1117 /* Update potential sum for this i atom from the interaction with this j atom. */
1118 velec = _mm_andnot_ps(dummy_mask,velec);
1119 velecsum = _mm_add_ps(velecsum,velec);
1123 fscal = _mm_andnot_ps(dummy_mask,fscal);
1125 /* Calculate temporary vectorial force */
1126 tx = _mm_mul_ps(fscal,dx32);
1127 ty = _mm_mul_ps(fscal,dy32);
1128 tz = _mm_mul_ps(fscal,dz32);
1130 /* Update vectorial force */
1131 fix3 = _mm_add_ps(fix3,tx);
1132 fiy3 = _mm_add_ps(fiy3,ty);
1133 fiz3 = _mm_add_ps(fiz3,tz);
1135 fjx2 = _mm_add_ps(fjx2,tx);
1136 fjy2 = _mm_add_ps(fjy2,ty);
1137 fjz2 = _mm_add_ps(fjz2,tz);
1139 /**************************
1140 * CALCULATE INTERACTIONS *
1141 **************************/
1143 r33 = _mm_mul_ps(rsq33,rinv33);
1144 r33 = _mm_andnot_ps(dummy_mask,r33);
1146 /* Calculate table index by multiplying r with table scale and truncate to integer */
1147 rt = _mm_mul_ps(r33,vftabscale);
1148 vfitab = _mm_cvttps_epi32(rt);
1149 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1150 vfitab = _mm_slli_epi32(vfitab,2);
1152 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1153 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1154 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1155 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1156 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1157 _MM_TRANSPOSE4_PS(Y,F,G,H);
1158 Heps = _mm_mul_ps(vfeps,H);
1159 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1160 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
1161 velec = _mm_mul_ps(qq33,VV);
1162 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
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 /* Calculate temporary vectorial force */
1174 tx = _mm_mul_ps(fscal,dx33);
1175 ty = _mm_mul_ps(fscal,dy33);
1176 tz = _mm_mul_ps(fscal,dz33);
1178 /* Update vectorial force */
1179 fix3 = _mm_add_ps(fix3,tx);
1180 fiy3 = _mm_add_ps(fiy3,ty);
1181 fiz3 = _mm_add_ps(fiz3,tz);
1183 fjx3 = _mm_add_ps(fjx3,tx);
1184 fjy3 = _mm_add_ps(fjy3,ty);
1185 fjz3 = _mm_add_ps(fjz3,tz);
1187 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1188 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1189 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1190 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1192 gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
1193 fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1195 /* Inner loop uses 396 flops */
1198 /* End of innermost loop */
1200 gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1201 f+i_coord_offset+DIM,fshift+i_shift_offset);
1204 /* Update potential energies */
1205 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1207 /* Increment number of inner iterations */
1208 inneriter += j_index_end - j_index_start;
1210 /* Outer loop uses 19 flops */
1213 /* Increment number of outer iterations */
1216 /* Update outer/inner flops */
1218 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_VF,outeriter*19 + inneriter*396);
1221 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwNone_GeomW4W4_F_sse2_single
1222 * Electrostatics interaction: CubicSplineTable
1223 * VdW interaction: None
1224 * Geometry: Water4-Water4
1225 * Calculate force/pot: Force
1228 nb_kernel_ElecCSTab_VdwNone_GeomW4W4_F_sse2_single
1229 (t_nblist * gmx_restrict nlist,
1230 rvec * gmx_restrict xx,
1231 rvec * gmx_restrict ff,
1232 t_forcerec * gmx_restrict fr,
1233 t_mdatoms * gmx_restrict mdatoms,
1234 nb_kernel_data_t * gmx_restrict kernel_data,
1235 t_nrnb * gmx_restrict nrnb)
1237 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1238 * just 0 for non-waters.
1239 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
1240 * jnr indices corresponding to data put in the four positions in the SIMD register.
1242 int i_shift_offset,i_coord_offset,outeriter,inneriter;
1243 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1244 int jnrA,jnrB,jnrC,jnrD;
1245 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1246 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1247 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
1248 real rcutoff_scalar;
1249 real *shiftvec,*fshift,*x,*f;
1250 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1251 real scratch[4*DIM];
1252 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1254 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1256 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1258 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1259 int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1260 __m128 jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1261 int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1262 __m128 jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1263 int vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
1264 __m128 jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1265 __m128 dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1266 __m128 dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1267 __m128 dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1268 __m128 dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1269 __m128 dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1270 __m128 dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1271 __m128 dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1272 __m128 dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1273 __m128 dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1274 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
1277 __m128i ifour = _mm_set1_epi32(4);
1278 __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
1280 __m128 dummy_mask,cutoff_mask;
1281 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
1282 __m128 one = _mm_set1_ps(1.0);
1283 __m128 two = _mm_set1_ps(2.0);
1289 jindex = nlist->jindex;
1291 shiftidx = nlist->shift;
1293 shiftvec = fr->shift_vec[0];
1294 fshift = fr->fshift[0];
1295 facel = _mm_set1_ps(fr->epsfac);
1296 charge = mdatoms->chargeA;
1298 vftab = kernel_data->table_elec->data;
1299 vftabscale = _mm_set1_ps(kernel_data->table_elec->scale);
1301 /* Setup water-specific parameters */
1302 inr = nlist->iinr[0];
1303 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
1304 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
1305 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
1307 jq1 = _mm_set1_ps(charge[inr+1]);
1308 jq2 = _mm_set1_ps(charge[inr+2]);
1309 jq3 = _mm_set1_ps(charge[inr+3]);
1310 qq11 = _mm_mul_ps(iq1,jq1);
1311 qq12 = _mm_mul_ps(iq1,jq2);
1312 qq13 = _mm_mul_ps(iq1,jq3);
1313 qq21 = _mm_mul_ps(iq2,jq1);
1314 qq22 = _mm_mul_ps(iq2,jq2);
1315 qq23 = _mm_mul_ps(iq2,jq3);
1316 qq31 = _mm_mul_ps(iq3,jq1);
1317 qq32 = _mm_mul_ps(iq3,jq2);
1318 qq33 = _mm_mul_ps(iq3,jq3);
1320 /* Avoid stupid compiler warnings */
1321 jnrA = jnrB = jnrC = jnrD = 0;
1322 j_coord_offsetA = 0;
1323 j_coord_offsetB = 0;
1324 j_coord_offsetC = 0;
1325 j_coord_offsetD = 0;
1330 for(iidx=0;iidx<4*DIM;iidx++)
1332 scratch[iidx] = 0.0;
1335 /* Start outer loop over neighborlists */
1336 for(iidx=0; iidx<nri; iidx++)
1338 /* Load shift vector for this list */
1339 i_shift_offset = DIM*shiftidx[iidx];
1341 /* Load limits for loop over neighbors */
1342 j_index_start = jindex[iidx];
1343 j_index_end = jindex[iidx+1];
1345 /* Get outer coordinate index */
1347 i_coord_offset = DIM*inr;
1349 /* Load i particle coords and add shift vector */
1350 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
1351 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1353 fix1 = _mm_setzero_ps();
1354 fiy1 = _mm_setzero_ps();
1355 fiz1 = _mm_setzero_ps();
1356 fix2 = _mm_setzero_ps();
1357 fiy2 = _mm_setzero_ps();
1358 fiz2 = _mm_setzero_ps();
1359 fix3 = _mm_setzero_ps();
1360 fiy3 = _mm_setzero_ps();
1361 fiz3 = _mm_setzero_ps();
1363 /* Start inner kernel loop */
1364 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1367 /* Get j neighbor index, and coordinate index */
1369 jnrB = jjnr[jidx+1];
1370 jnrC = jjnr[jidx+2];
1371 jnrD = jjnr[jidx+3];
1372 j_coord_offsetA = DIM*jnrA;
1373 j_coord_offsetB = DIM*jnrB;
1374 j_coord_offsetC = DIM*jnrC;
1375 j_coord_offsetD = DIM*jnrD;
1377 /* load j atom coordinates */
1378 gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
1379 x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
1380 &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1382 /* Calculate displacement vector */
1383 dx11 = _mm_sub_ps(ix1,jx1);
1384 dy11 = _mm_sub_ps(iy1,jy1);
1385 dz11 = _mm_sub_ps(iz1,jz1);
1386 dx12 = _mm_sub_ps(ix1,jx2);
1387 dy12 = _mm_sub_ps(iy1,jy2);
1388 dz12 = _mm_sub_ps(iz1,jz2);
1389 dx13 = _mm_sub_ps(ix1,jx3);
1390 dy13 = _mm_sub_ps(iy1,jy3);
1391 dz13 = _mm_sub_ps(iz1,jz3);
1392 dx21 = _mm_sub_ps(ix2,jx1);
1393 dy21 = _mm_sub_ps(iy2,jy1);
1394 dz21 = _mm_sub_ps(iz2,jz1);
1395 dx22 = _mm_sub_ps(ix2,jx2);
1396 dy22 = _mm_sub_ps(iy2,jy2);
1397 dz22 = _mm_sub_ps(iz2,jz2);
1398 dx23 = _mm_sub_ps(ix2,jx3);
1399 dy23 = _mm_sub_ps(iy2,jy3);
1400 dz23 = _mm_sub_ps(iz2,jz3);
1401 dx31 = _mm_sub_ps(ix3,jx1);
1402 dy31 = _mm_sub_ps(iy3,jy1);
1403 dz31 = _mm_sub_ps(iz3,jz1);
1404 dx32 = _mm_sub_ps(ix3,jx2);
1405 dy32 = _mm_sub_ps(iy3,jy2);
1406 dz32 = _mm_sub_ps(iz3,jz2);
1407 dx33 = _mm_sub_ps(ix3,jx3);
1408 dy33 = _mm_sub_ps(iy3,jy3);
1409 dz33 = _mm_sub_ps(iz3,jz3);
1411 /* Calculate squared distance and things based on it */
1412 rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1413 rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1414 rsq13 = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
1415 rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1416 rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1417 rsq23 = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
1418 rsq31 = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
1419 rsq32 = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
1420 rsq33 = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
1422 rinv11 = gmx_mm_invsqrt_ps(rsq11);
1423 rinv12 = gmx_mm_invsqrt_ps(rsq12);
1424 rinv13 = gmx_mm_invsqrt_ps(rsq13);
1425 rinv21 = gmx_mm_invsqrt_ps(rsq21);
1426 rinv22 = gmx_mm_invsqrt_ps(rsq22);
1427 rinv23 = gmx_mm_invsqrt_ps(rsq23);
1428 rinv31 = gmx_mm_invsqrt_ps(rsq31);
1429 rinv32 = gmx_mm_invsqrt_ps(rsq32);
1430 rinv33 = gmx_mm_invsqrt_ps(rsq33);
1432 fjx1 = _mm_setzero_ps();
1433 fjy1 = _mm_setzero_ps();
1434 fjz1 = _mm_setzero_ps();
1435 fjx2 = _mm_setzero_ps();
1436 fjy2 = _mm_setzero_ps();
1437 fjz2 = _mm_setzero_ps();
1438 fjx3 = _mm_setzero_ps();
1439 fjy3 = _mm_setzero_ps();
1440 fjz3 = _mm_setzero_ps();
1442 /**************************
1443 * CALCULATE INTERACTIONS *
1444 **************************/
1446 r11 = _mm_mul_ps(rsq11,rinv11);
1448 /* Calculate table index by multiplying r with table scale and truncate to integer */
1449 rt = _mm_mul_ps(r11,vftabscale);
1450 vfitab = _mm_cvttps_epi32(rt);
1451 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1452 vfitab = _mm_slli_epi32(vfitab,2);
1454 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1455 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1456 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1457 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1458 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1459 _MM_TRANSPOSE4_PS(Y,F,G,H);
1460 Heps = _mm_mul_ps(vfeps,H);
1461 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1462 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1463 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq11,FF),_mm_mul_ps(vftabscale,rinv11)));
1467 /* Calculate temporary vectorial force */
1468 tx = _mm_mul_ps(fscal,dx11);
1469 ty = _mm_mul_ps(fscal,dy11);
1470 tz = _mm_mul_ps(fscal,dz11);
1472 /* Update vectorial force */
1473 fix1 = _mm_add_ps(fix1,tx);
1474 fiy1 = _mm_add_ps(fiy1,ty);
1475 fiz1 = _mm_add_ps(fiz1,tz);
1477 fjx1 = _mm_add_ps(fjx1,tx);
1478 fjy1 = _mm_add_ps(fjy1,ty);
1479 fjz1 = _mm_add_ps(fjz1,tz);
1481 /**************************
1482 * CALCULATE INTERACTIONS *
1483 **************************/
1485 r12 = _mm_mul_ps(rsq12,rinv12);
1487 /* Calculate table index by multiplying r with table scale and truncate to integer */
1488 rt = _mm_mul_ps(r12,vftabscale);
1489 vfitab = _mm_cvttps_epi32(rt);
1490 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1491 vfitab = _mm_slli_epi32(vfitab,2);
1493 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1494 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1495 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1496 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1497 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1498 _MM_TRANSPOSE4_PS(Y,F,G,H);
1499 Heps = _mm_mul_ps(vfeps,H);
1500 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1501 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1502 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq12,FF),_mm_mul_ps(vftabscale,rinv12)));
1506 /* Calculate temporary vectorial force */
1507 tx = _mm_mul_ps(fscal,dx12);
1508 ty = _mm_mul_ps(fscal,dy12);
1509 tz = _mm_mul_ps(fscal,dz12);
1511 /* Update vectorial force */
1512 fix1 = _mm_add_ps(fix1,tx);
1513 fiy1 = _mm_add_ps(fiy1,ty);
1514 fiz1 = _mm_add_ps(fiz1,tz);
1516 fjx2 = _mm_add_ps(fjx2,tx);
1517 fjy2 = _mm_add_ps(fjy2,ty);
1518 fjz2 = _mm_add_ps(fjz2,tz);
1520 /**************************
1521 * CALCULATE INTERACTIONS *
1522 **************************/
1524 r13 = _mm_mul_ps(rsq13,rinv13);
1526 /* Calculate table index by multiplying r with table scale and truncate to integer */
1527 rt = _mm_mul_ps(r13,vftabscale);
1528 vfitab = _mm_cvttps_epi32(rt);
1529 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1530 vfitab = _mm_slli_epi32(vfitab,2);
1532 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1533 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1534 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1535 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1536 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1537 _MM_TRANSPOSE4_PS(Y,F,G,H);
1538 Heps = _mm_mul_ps(vfeps,H);
1539 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1540 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1541 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq13,FF),_mm_mul_ps(vftabscale,rinv13)));
1545 /* Calculate temporary vectorial force */
1546 tx = _mm_mul_ps(fscal,dx13);
1547 ty = _mm_mul_ps(fscal,dy13);
1548 tz = _mm_mul_ps(fscal,dz13);
1550 /* Update vectorial force */
1551 fix1 = _mm_add_ps(fix1,tx);
1552 fiy1 = _mm_add_ps(fiy1,ty);
1553 fiz1 = _mm_add_ps(fiz1,tz);
1555 fjx3 = _mm_add_ps(fjx3,tx);
1556 fjy3 = _mm_add_ps(fjy3,ty);
1557 fjz3 = _mm_add_ps(fjz3,tz);
1559 /**************************
1560 * CALCULATE INTERACTIONS *
1561 **************************/
1563 r21 = _mm_mul_ps(rsq21,rinv21);
1565 /* Calculate table index by multiplying r with table scale and truncate to integer */
1566 rt = _mm_mul_ps(r21,vftabscale);
1567 vfitab = _mm_cvttps_epi32(rt);
1568 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1569 vfitab = _mm_slli_epi32(vfitab,2);
1571 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1572 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1573 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1574 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1575 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1576 _MM_TRANSPOSE4_PS(Y,F,G,H);
1577 Heps = _mm_mul_ps(vfeps,H);
1578 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1579 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1580 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq21,FF),_mm_mul_ps(vftabscale,rinv21)));
1584 /* Calculate temporary vectorial force */
1585 tx = _mm_mul_ps(fscal,dx21);
1586 ty = _mm_mul_ps(fscal,dy21);
1587 tz = _mm_mul_ps(fscal,dz21);
1589 /* Update vectorial force */
1590 fix2 = _mm_add_ps(fix2,tx);
1591 fiy2 = _mm_add_ps(fiy2,ty);
1592 fiz2 = _mm_add_ps(fiz2,tz);
1594 fjx1 = _mm_add_ps(fjx1,tx);
1595 fjy1 = _mm_add_ps(fjy1,ty);
1596 fjz1 = _mm_add_ps(fjz1,tz);
1598 /**************************
1599 * CALCULATE INTERACTIONS *
1600 **************************/
1602 r22 = _mm_mul_ps(rsq22,rinv22);
1604 /* Calculate table index by multiplying r with table scale and truncate to integer */
1605 rt = _mm_mul_ps(r22,vftabscale);
1606 vfitab = _mm_cvttps_epi32(rt);
1607 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1608 vfitab = _mm_slli_epi32(vfitab,2);
1610 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1611 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1612 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1613 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1614 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1615 _MM_TRANSPOSE4_PS(Y,F,G,H);
1616 Heps = _mm_mul_ps(vfeps,H);
1617 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1618 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1619 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq22,FF),_mm_mul_ps(vftabscale,rinv22)));
1623 /* Calculate temporary vectorial force */
1624 tx = _mm_mul_ps(fscal,dx22);
1625 ty = _mm_mul_ps(fscal,dy22);
1626 tz = _mm_mul_ps(fscal,dz22);
1628 /* Update vectorial force */
1629 fix2 = _mm_add_ps(fix2,tx);
1630 fiy2 = _mm_add_ps(fiy2,ty);
1631 fiz2 = _mm_add_ps(fiz2,tz);
1633 fjx2 = _mm_add_ps(fjx2,tx);
1634 fjy2 = _mm_add_ps(fjy2,ty);
1635 fjz2 = _mm_add_ps(fjz2,tz);
1637 /**************************
1638 * CALCULATE INTERACTIONS *
1639 **************************/
1641 r23 = _mm_mul_ps(rsq23,rinv23);
1643 /* Calculate table index by multiplying r with table scale and truncate to integer */
1644 rt = _mm_mul_ps(r23,vftabscale);
1645 vfitab = _mm_cvttps_epi32(rt);
1646 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1647 vfitab = _mm_slli_epi32(vfitab,2);
1649 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1650 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1651 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1652 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1653 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1654 _MM_TRANSPOSE4_PS(Y,F,G,H);
1655 Heps = _mm_mul_ps(vfeps,H);
1656 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1657 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1658 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq23,FF),_mm_mul_ps(vftabscale,rinv23)));
1662 /* Calculate temporary vectorial force */
1663 tx = _mm_mul_ps(fscal,dx23);
1664 ty = _mm_mul_ps(fscal,dy23);
1665 tz = _mm_mul_ps(fscal,dz23);
1667 /* Update vectorial force */
1668 fix2 = _mm_add_ps(fix2,tx);
1669 fiy2 = _mm_add_ps(fiy2,ty);
1670 fiz2 = _mm_add_ps(fiz2,tz);
1672 fjx3 = _mm_add_ps(fjx3,tx);
1673 fjy3 = _mm_add_ps(fjy3,ty);
1674 fjz3 = _mm_add_ps(fjz3,tz);
1676 /**************************
1677 * CALCULATE INTERACTIONS *
1678 **************************/
1680 r31 = _mm_mul_ps(rsq31,rinv31);
1682 /* Calculate table index by multiplying r with table scale and truncate to integer */
1683 rt = _mm_mul_ps(r31,vftabscale);
1684 vfitab = _mm_cvttps_epi32(rt);
1685 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1686 vfitab = _mm_slli_epi32(vfitab,2);
1688 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1689 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1690 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1691 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1692 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1693 _MM_TRANSPOSE4_PS(Y,F,G,H);
1694 Heps = _mm_mul_ps(vfeps,H);
1695 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1696 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1697 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq31,FF),_mm_mul_ps(vftabscale,rinv31)));
1701 /* Calculate temporary vectorial force */
1702 tx = _mm_mul_ps(fscal,dx31);
1703 ty = _mm_mul_ps(fscal,dy31);
1704 tz = _mm_mul_ps(fscal,dz31);
1706 /* Update vectorial force */
1707 fix3 = _mm_add_ps(fix3,tx);
1708 fiy3 = _mm_add_ps(fiy3,ty);
1709 fiz3 = _mm_add_ps(fiz3,tz);
1711 fjx1 = _mm_add_ps(fjx1,tx);
1712 fjy1 = _mm_add_ps(fjy1,ty);
1713 fjz1 = _mm_add_ps(fjz1,tz);
1715 /**************************
1716 * CALCULATE INTERACTIONS *
1717 **************************/
1719 r32 = _mm_mul_ps(rsq32,rinv32);
1721 /* Calculate table index by multiplying r with table scale and truncate to integer */
1722 rt = _mm_mul_ps(r32,vftabscale);
1723 vfitab = _mm_cvttps_epi32(rt);
1724 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1725 vfitab = _mm_slli_epi32(vfitab,2);
1727 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1728 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1729 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1730 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1731 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1732 _MM_TRANSPOSE4_PS(Y,F,G,H);
1733 Heps = _mm_mul_ps(vfeps,H);
1734 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1735 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1736 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq32,FF),_mm_mul_ps(vftabscale,rinv32)));
1740 /* Calculate temporary vectorial force */
1741 tx = _mm_mul_ps(fscal,dx32);
1742 ty = _mm_mul_ps(fscal,dy32);
1743 tz = _mm_mul_ps(fscal,dz32);
1745 /* Update vectorial force */
1746 fix3 = _mm_add_ps(fix3,tx);
1747 fiy3 = _mm_add_ps(fiy3,ty);
1748 fiz3 = _mm_add_ps(fiz3,tz);
1750 fjx2 = _mm_add_ps(fjx2,tx);
1751 fjy2 = _mm_add_ps(fjy2,ty);
1752 fjz2 = _mm_add_ps(fjz2,tz);
1754 /**************************
1755 * CALCULATE INTERACTIONS *
1756 **************************/
1758 r33 = _mm_mul_ps(rsq33,rinv33);
1760 /* Calculate table index by multiplying r with table scale and truncate to integer */
1761 rt = _mm_mul_ps(r33,vftabscale);
1762 vfitab = _mm_cvttps_epi32(rt);
1763 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1764 vfitab = _mm_slli_epi32(vfitab,2);
1766 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1767 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1768 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1769 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1770 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1771 _MM_TRANSPOSE4_PS(Y,F,G,H);
1772 Heps = _mm_mul_ps(vfeps,H);
1773 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1774 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1775 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq33,FF),_mm_mul_ps(vftabscale,rinv33)));
1779 /* Calculate temporary vectorial force */
1780 tx = _mm_mul_ps(fscal,dx33);
1781 ty = _mm_mul_ps(fscal,dy33);
1782 tz = _mm_mul_ps(fscal,dz33);
1784 /* Update vectorial force */
1785 fix3 = _mm_add_ps(fix3,tx);
1786 fiy3 = _mm_add_ps(fiy3,ty);
1787 fiz3 = _mm_add_ps(fiz3,tz);
1789 fjx3 = _mm_add_ps(fjx3,tx);
1790 fjy3 = _mm_add_ps(fjy3,ty);
1791 fjz3 = _mm_add_ps(fjz3,tz);
1793 fjptrA = f+j_coord_offsetA;
1794 fjptrB = f+j_coord_offsetB;
1795 fjptrC = f+j_coord_offsetC;
1796 fjptrD = f+j_coord_offsetD;
1798 gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
1799 fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1801 /* Inner loop uses 351 flops */
1804 if(jidx<j_index_end)
1807 /* Get j neighbor index, and coordinate index */
1808 jnrlistA = jjnr[jidx];
1809 jnrlistB = jjnr[jidx+1];
1810 jnrlistC = jjnr[jidx+2];
1811 jnrlistD = jjnr[jidx+3];
1812 /* Sign of each element will be negative for non-real atoms.
1813 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1814 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1816 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1817 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
1818 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
1819 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
1820 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
1821 j_coord_offsetA = DIM*jnrA;
1822 j_coord_offsetB = DIM*jnrB;
1823 j_coord_offsetC = DIM*jnrC;
1824 j_coord_offsetD = DIM*jnrD;
1826 /* load j atom coordinates */
1827 gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
1828 x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
1829 &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1831 /* Calculate displacement vector */
1832 dx11 = _mm_sub_ps(ix1,jx1);
1833 dy11 = _mm_sub_ps(iy1,jy1);
1834 dz11 = _mm_sub_ps(iz1,jz1);
1835 dx12 = _mm_sub_ps(ix1,jx2);
1836 dy12 = _mm_sub_ps(iy1,jy2);
1837 dz12 = _mm_sub_ps(iz1,jz2);
1838 dx13 = _mm_sub_ps(ix1,jx3);
1839 dy13 = _mm_sub_ps(iy1,jy3);
1840 dz13 = _mm_sub_ps(iz1,jz3);
1841 dx21 = _mm_sub_ps(ix2,jx1);
1842 dy21 = _mm_sub_ps(iy2,jy1);
1843 dz21 = _mm_sub_ps(iz2,jz1);
1844 dx22 = _mm_sub_ps(ix2,jx2);
1845 dy22 = _mm_sub_ps(iy2,jy2);
1846 dz22 = _mm_sub_ps(iz2,jz2);
1847 dx23 = _mm_sub_ps(ix2,jx3);
1848 dy23 = _mm_sub_ps(iy2,jy3);
1849 dz23 = _mm_sub_ps(iz2,jz3);
1850 dx31 = _mm_sub_ps(ix3,jx1);
1851 dy31 = _mm_sub_ps(iy3,jy1);
1852 dz31 = _mm_sub_ps(iz3,jz1);
1853 dx32 = _mm_sub_ps(ix3,jx2);
1854 dy32 = _mm_sub_ps(iy3,jy2);
1855 dz32 = _mm_sub_ps(iz3,jz2);
1856 dx33 = _mm_sub_ps(ix3,jx3);
1857 dy33 = _mm_sub_ps(iy3,jy3);
1858 dz33 = _mm_sub_ps(iz3,jz3);
1860 /* Calculate squared distance and things based on it */
1861 rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1862 rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1863 rsq13 = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
1864 rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1865 rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1866 rsq23 = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
1867 rsq31 = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
1868 rsq32 = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
1869 rsq33 = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
1871 rinv11 = gmx_mm_invsqrt_ps(rsq11);
1872 rinv12 = gmx_mm_invsqrt_ps(rsq12);
1873 rinv13 = gmx_mm_invsqrt_ps(rsq13);
1874 rinv21 = gmx_mm_invsqrt_ps(rsq21);
1875 rinv22 = gmx_mm_invsqrt_ps(rsq22);
1876 rinv23 = gmx_mm_invsqrt_ps(rsq23);
1877 rinv31 = gmx_mm_invsqrt_ps(rsq31);
1878 rinv32 = gmx_mm_invsqrt_ps(rsq32);
1879 rinv33 = gmx_mm_invsqrt_ps(rsq33);
1881 fjx1 = _mm_setzero_ps();
1882 fjy1 = _mm_setzero_ps();
1883 fjz1 = _mm_setzero_ps();
1884 fjx2 = _mm_setzero_ps();
1885 fjy2 = _mm_setzero_ps();
1886 fjz2 = _mm_setzero_ps();
1887 fjx3 = _mm_setzero_ps();
1888 fjy3 = _mm_setzero_ps();
1889 fjz3 = _mm_setzero_ps();
1891 /**************************
1892 * CALCULATE INTERACTIONS *
1893 **************************/
1895 r11 = _mm_mul_ps(rsq11,rinv11);
1896 r11 = _mm_andnot_ps(dummy_mask,r11);
1898 /* Calculate table index by multiplying r with table scale and truncate to integer */
1899 rt = _mm_mul_ps(r11,vftabscale);
1900 vfitab = _mm_cvttps_epi32(rt);
1901 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1902 vfitab = _mm_slli_epi32(vfitab,2);
1904 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1905 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1906 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1907 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1908 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1909 _MM_TRANSPOSE4_PS(Y,F,G,H);
1910 Heps = _mm_mul_ps(vfeps,H);
1911 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1912 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1913 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq11,FF),_mm_mul_ps(vftabscale,rinv11)));
1917 fscal = _mm_andnot_ps(dummy_mask,fscal);
1919 /* Calculate temporary vectorial force */
1920 tx = _mm_mul_ps(fscal,dx11);
1921 ty = _mm_mul_ps(fscal,dy11);
1922 tz = _mm_mul_ps(fscal,dz11);
1924 /* Update vectorial force */
1925 fix1 = _mm_add_ps(fix1,tx);
1926 fiy1 = _mm_add_ps(fiy1,ty);
1927 fiz1 = _mm_add_ps(fiz1,tz);
1929 fjx1 = _mm_add_ps(fjx1,tx);
1930 fjy1 = _mm_add_ps(fjy1,ty);
1931 fjz1 = _mm_add_ps(fjz1,tz);
1933 /**************************
1934 * CALCULATE INTERACTIONS *
1935 **************************/
1937 r12 = _mm_mul_ps(rsq12,rinv12);
1938 r12 = _mm_andnot_ps(dummy_mask,r12);
1940 /* Calculate table index by multiplying r with table scale and truncate to integer */
1941 rt = _mm_mul_ps(r12,vftabscale);
1942 vfitab = _mm_cvttps_epi32(rt);
1943 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1944 vfitab = _mm_slli_epi32(vfitab,2);
1946 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1947 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1948 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1949 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1950 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1951 _MM_TRANSPOSE4_PS(Y,F,G,H);
1952 Heps = _mm_mul_ps(vfeps,H);
1953 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1954 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1955 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq12,FF),_mm_mul_ps(vftabscale,rinv12)));
1959 fscal = _mm_andnot_ps(dummy_mask,fscal);
1961 /* Calculate temporary vectorial force */
1962 tx = _mm_mul_ps(fscal,dx12);
1963 ty = _mm_mul_ps(fscal,dy12);
1964 tz = _mm_mul_ps(fscal,dz12);
1966 /* Update vectorial force */
1967 fix1 = _mm_add_ps(fix1,tx);
1968 fiy1 = _mm_add_ps(fiy1,ty);
1969 fiz1 = _mm_add_ps(fiz1,tz);
1971 fjx2 = _mm_add_ps(fjx2,tx);
1972 fjy2 = _mm_add_ps(fjy2,ty);
1973 fjz2 = _mm_add_ps(fjz2,tz);
1975 /**************************
1976 * CALCULATE INTERACTIONS *
1977 **************************/
1979 r13 = _mm_mul_ps(rsq13,rinv13);
1980 r13 = _mm_andnot_ps(dummy_mask,r13);
1982 /* Calculate table index by multiplying r with table scale and truncate to integer */
1983 rt = _mm_mul_ps(r13,vftabscale);
1984 vfitab = _mm_cvttps_epi32(rt);
1985 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1986 vfitab = _mm_slli_epi32(vfitab,2);
1988 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1989 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1990 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1991 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1992 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1993 _MM_TRANSPOSE4_PS(Y,F,G,H);
1994 Heps = _mm_mul_ps(vfeps,H);
1995 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1996 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1997 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq13,FF),_mm_mul_ps(vftabscale,rinv13)));
2001 fscal = _mm_andnot_ps(dummy_mask,fscal);
2003 /* Calculate temporary vectorial force */
2004 tx = _mm_mul_ps(fscal,dx13);
2005 ty = _mm_mul_ps(fscal,dy13);
2006 tz = _mm_mul_ps(fscal,dz13);
2008 /* Update vectorial force */
2009 fix1 = _mm_add_ps(fix1,tx);
2010 fiy1 = _mm_add_ps(fiy1,ty);
2011 fiz1 = _mm_add_ps(fiz1,tz);
2013 fjx3 = _mm_add_ps(fjx3,tx);
2014 fjy3 = _mm_add_ps(fjy3,ty);
2015 fjz3 = _mm_add_ps(fjz3,tz);
2017 /**************************
2018 * CALCULATE INTERACTIONS *
2019 **************************/
2021 r21 = _mm_mul_ps(rsq21,rinv21);
2022 r21 = _mm_andnot_ps(dummy_mask,r21);
2024 /* Calculate table index by multiplying r with table scale and truncate to integer */
2025 rt = _mm_mul_ps(r21,vftabscale);
2026 vfitab = _mm_cvttps_epi32(rt);
2027 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
2028 vfitab = _mm_slli_epi32(vfitab,2);
2030 /* CUBIC SPLINE TABLE ELECTROSTATICS */
2031 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
2032 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
2033 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
2034 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
2035 _MM_TRANSPOSE4_PS(Y,F,G,H);
2036 Heps = _mm_mul_ps(vfeps,H);
2037 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
2038 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
2039 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq21,FF),_mm_mul_ps(vftabscale,rinv21)));
2043 fscal = _mm_andnot_ps(dummy_mask,fscal);
2045 /* Calculate temporary vectorial force */
2046 tx = _mm_mul_ps(fscal,dx21);
2047 ty = _mm_mul_ps(fscal,dy21);
2048 tz = _mm_mul_ps(fscal,dz21);
2050 /* Update vectorial force */
2051 fix2 = _mm_add_ps(fix2,tx);
2052 fiy2 = _mm_add_ps(fiy2,ty);
2053 fiz2 = _mm_add_ps(fiz2,tz);
2055 fjx1 = _mm_add_ps(fjx1,tx);
2056 fjy1 = _mm_add_ps(fjy1,ty);
2057 fjz1 = _mm_add_ps(fjz1,tz);
2059 /**************************
2060 * CALCULATE INTERACTIONS *
2061 **************************/
2063 r22 = _mm_mul_ps(rsq22,rinv22);
2064 r22 = _mm_andnot_ps(dummy_mask,r22);
2066 /* Calculate table index by multiplying r with table scale and truncate to integer */
2067 rt = _mm_mul_ps(r22,vftabscale);
2068 vfitab = _mm_cvttps_epi32(rt);
2069 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
2070 vfitab = _mm_slli_epi32(vfitab,2);
2072 /* CUBIC SPLINE TABLE ELECTROSTATICS */
2073 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
2074 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
2075 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
2076 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
2077 _MM_TRANSPOSE4_PS(Y,F,G,H);
2078 Heps = _mm_mul_ps(vfeps,H);
2079 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
2080 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
2081 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq22,FF),_mm_mul_ps(vftabscale,rinv22)));
2085 fscal = _mm_andnot_ps(dummy_mask,fscal);
2087 /* Calculate temporary vectorial force */
2088 tx = _mm_mul_ps(fscal,dx22);
2089 ty = _mm_mul_ps(fscal,dy22);
2090 tz = _mm_mul_ps(fscal,dz22);
2092 /* Update vectorial force */
2093 fix2 = _mm_add_ps(fix2,tx);
2094 fiy2 = _mm_add_ps(fiy2,ty);
2095 fiz2 = _mm_add_ps(fiz2,tz);
2097 fjx2 = _mm_add_ps(fjx2,tx);
2098 fjy2 = _mm_add_ps(fjy2,ty);
2099 fjz2 = _mm_add_ps(fjz2,tz);
2101 /**************************
2102 * CALCULATE INTERACTIONS *
2103 **************************/
2105 r23 = _mm_mul_ps(rsq23,rinv23);
2106 r23 = _mm_andnot_ps(dummy_mask,r23);
2108 /* Calculate table index by multiplying r with table scale and truncate to integer */
2109 rt = _mm_mul_ps(r23,vftabscale);
2110 vfitab = _mm_cvttps_epi32(rt);
2111 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
2112 vfitab = _mm_slli_epi32(vfitab,2);
2114 /* CUBIC SPLINE TABLE ELECTROSTATICS */
2115 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
2116 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
2117 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
2118 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
2119 _MM_TRANSPOSE4_PS(Y,F,G,H);
2120 Heps = _mm_mul_ps(vfeps,H);
2121 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
2122 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
2123 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq23,FF),_mm_mul_ps(vftabscale,rinv23)));
2127 fscal = _mm_andnot_ps(dummy_mask,fscal);
2129 /* Calculate temporary vectorial force */
2130 tx = _mm_mul_ps(fscal,dx23);
2131 ty = _mm_mul_ps(fscal,dy23);
2132 tz = _mm_mul_ps(fscal,dz23);
2134 /* Update vectorial force */
2135 fix2 = _mm_add_ps(fix2,tx);
2136 fiy2 = _mm_add_ps(fiy2,ty);
2137 fiz2 = _mm_add_ps(fiz2,tz);
2139 fjx3 = _mm_add_ps(fjx3,tx);
2140 fjy3 = _mm_add_ps(fjy3,ty);
2141 fjz3 = _mm_add_ps(fjz3,tz);
2143 /**************************
2144 * CALCULATE INTERACTIONS *
2145 **************************/
2147 r31 = _mm_mul_ps(rsq31,rinv31);
2148 r31 = _mm_andnot_ps(dummy_mask,r31);
2150 /* Calculate table index by multiplying r with table scale and truncate to integer */
2151 rt = _mm_mul_ps(r31,vftabscale);
2152 vfitab = _mm_cvttps_epi32(rt);
2153 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
2154 vfitab = _mm_slli_epi32(vfitab,2);
2156 /* CUBIC SPLINE TABLE ELECTROSTATICS */
2157 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
2158 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
2159 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
2160 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
2161 _MM_TRANSPOSE4_PS(Y,F,G,H);
2162 Heps = _mm_mul_ps(vfeps,H);
2163 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
2164 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
2165 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq31,FF),_mm_mul_ps(vftabscale,rinv31)));
2169 fscal = _mm_andnot_ps(dummy_mask,fscal);
2171 /* Calculate temporary vectorial force */
2172 tx = _mm_mul_ps(fscal,dx31);
2173 ty = _mm_mul_ps(fscal,dy31);
2174 tz = _mm_mul_ps(fscal,dz31);
2176 /* Update vectorial force */
2177 fix3 = _mm_add_ps(fix3,tx);
2178 fiy3 = _mm_add_ps(fiy3,ty);
2179 fiz3 = _mm_add_ps(fiz3,tz);
2181 fjx1 = _mm_add_ps(fjx1,tx);
2182 fjy1 = _mm_add_ps(fjy1,ty);
2183 fjz1 = _mm_add_ps(fjz1,tz);
2185 /**************************
2186 * CALCULATE INTERACTIONS *
2187 **************************/
2189 r32 = _mm_mul_ps(rsq32,rinv32);
2190 r32 = _mm_andnot_ps(dummy_mask,r32);
2192 /* Calculate table index by multiplying r with table scale and truncate to integer */
2193 rt = _mm_mul_ps(r32,vftabscale);
2194 vfitab = _mm_cvttps_epi32(rt);
2195 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
2196 vfitab = _mm_slli_epi32(vfitab,2);
2198 /* CUBIC SPLINE TABLE ELECTROSTATICS */
2199 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
2200 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
2201 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
2202 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
2203 _MM_TRANSPOSE4_PS(Y,F,G,H);
2204 Heps = _mm_mul_ps(vfeps,H);
2205 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
2206 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
2207 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq32,FF),_mm_mul_ps(vftabscale,rinv32)));
2211 fscal = _mm_andnot_ps(dummy_mask,fscal);
2213 /* Calculate temporary vectorial force */
2214 tx = _mm_mul_ps(fscal,dx32);
2215 ty = _mm_mul_ps(fscal,dy32);
2216 tz = _mm_mul_ps(fscal,dz32);
2218 /* Update vectorial force */
2219 fix3 = _mm_add_ps(fix3,tx);
2220 fiy3 = _mm_add_ps(fiy3,ty);
2221 fiz3 = _mm_add_ps(fiz3,tz);
2223 fjx2 = _mm_add_ps(fjx2,tx);
2224 fjy2 = _mm_add_ps(fjy2,ty);
2225 fjz2 = _mm_add_ps(fjz2,tz);
2227 /**************************
2228 * CALCULATE INTERACTIONS *
2229 **************************/
2231 r33 = _mm_mul_ps(rsq33,rinv33);
2232 r33 = _mm_andnot_ps(dummy_mask,r33);
2234 /* Calculate table index by multiplying r with table scale and truncate to integer */
2235 rt = _mm_mul_ps(r33,vftabscale);
2236 vfitab = _mm_cvttps_epi32(rt);
2237 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
2238 vfitab = _mm_slli_epi32(vfitab,2);
2240 /* CUBIC SPLINE TABLE ELECTROSTATICS */
2241 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
2242 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
2243 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
2244 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
2245 _MM_TRANSPOSE4_PS(Y,F,G,H);
2246 Heps = _mm_mul_ps(vfeps,H);
2247 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
2248 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
2249 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq33,FF),_mm_mul_ps(vftabscale,rinv33)));
2253 fscal = _mm_andnot_ps(dummy_mask,fscal);
2255 /* Calculate temporary vectorial force */
2256 tx = _mm_mul_ps(fscal,dx33);
2257 ty = _mm_mul_ps(fscal,dy33);
2258 tz = _mm_mul_ps(fscal,dz33);
2260 /* Update vectorial force */
2261 fix3 = _mm_add_ps(fix3,tx);
2262 fiy3 = _mm_add_ps(fiy3,ty);
2263 fiz3 = _mm_add_ps(fiz3,tz);
2265 fjx3 = _mm_add_ps(fjx3,tx);
2266 fjy3 = _mm_add_ps(fjy3,ty);
2267 fjz3 = _mm_add_ps(fjz3,tz);
2269 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2270 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2271 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2272 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2274 gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
2275 fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2277 /* Inner loop uses 360 flops */
2280 /* End of innermost loop */
2282 gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2283 f+i_coord_offset+DIM,fshift+i_shift_offset);
2285 /* Increment number of inner iterations */
2286 inneriter += j_index_end - j_index_start;
2288 /* Outer loop uses 18 flops */
2291 /* Increment number of outer iterations */
2294 /* Update outer/inner flops */
2296 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_F,outeriter*18 + inneriter*360);