File: | gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecRFCut_VdwCSTab_GeomW3W3_sse4_1_single.c |
Location: | line 1284, column 22 |
Description: | Value stored to 'one_sixth' during its initialization is never read |
1 | /* |
2 | * This file is part of the GROMACS molecular simulation package. |
3 | * |
4 | * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by |
5 | * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl, |
6 | * and including many others, as listed in the AUTHORS file in the |
7 | * top-level source directory and at http://www.gromacs.org. |
8 | * |
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. |
13 | * |
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. |
18 | * |
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. |
23 | * |
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. |
31 | * |
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. |
34 | */ |
35 | /* |
36 | * Note: this file was generated by the GROMACS sse4_1_single kernel generator. |
37 | */ |
38 | #ifdef HAVE_CONFIG_H1 |
39 | #include <config.h> |
40 | #endif |
41 | |
42 | #include <math.h> |
43 | |
44 | #include "../nb_kernel.h" |
45 | #include "types/simple.h" |
46 | #include "gromacs/math/vec.h" |
47 | #include "nrnb.h" |
48 | |
49 | #include "gromacs/simd/math_x86_sse4_1_single.h" |
50 | #include "kernelutil_x86_sse4_1_single.h" |
51 | |
52 | /* |
53 | * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwCSTab_GeomW3W3_VF_sse4_1_single |
54 | * Electrostatics interaction: ReactionField |
55 | * VdW interaction: CubicSplineTable |
56 | * Geometry: Water3-Water3 |
57 | * Calculate force/pot: PotentialAndForce |
58 | */ |
59 | void |
60 | nb_kernel_ElecRFCut_VdwCSTab_GeomW3W3_VF_sse4_1_single |
61 | (t_nblist * gmx_restrict nlist, |
62 | rvec * gmx_restrict xx, |
63 | rvec * gmx_restrict ff, |
64 | t_forcerec * gmx_restrict fr, |
65 | t_mdatoms * gmx_restrict mdatoms, |
66 | nb_kernel_data_t gmx_unused__attribute__ ((unused)) * gmx_restrict kernel_data, |
67 | t_nrnb * gmx_restrict nrnb) |
68 | { |
69 | /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or |
70 | * just 0 for non-waters. |
71 | * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different |
72 | * jnr indices corresponding to data put in the four positions in the SIMD register. |
73 | */ |
74 | int i_shift_offset,i_coord_offset,outeriter,inneriter; |
75 | int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx; |
76 | int jnrA,jnrB,jnrC,jnrD; |
77 | int jnrlistA,jnrlistB,jnrlistC,jnrlistD; |
78 | int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD; |
79 | int *iinr,*jindex,*jjnr,*shiftidx,*gid; |
80 | real rcutoff_scalar; |
81 | real *shiftvec,*fshift,*x,*f; |
82 | real *fjptrA,*fjptrB,*fjptrC,*fjptrD; |
83 | real scratch[4*DIM3]; |
84 | __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall; |
85 | int vdwioffset0; |
86 | __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0; |
87 | int vdwioffset1; |
88 | __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1; |
89 | int vdwioffset2; |
90 | __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2; |
91 | int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D; |
92 | __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0; |
93 | int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D; |
94 | __m128 jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1; |
95 | int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D; |
96 | __m128 jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2; |
97 | __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00; |
98 | __m128 dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01; |
99 | __m128 dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02; |
100 | __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10; |
101 | __m128 dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11; |
102 | __m128 dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12; |
103 | __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20; |
104 | __m128 dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21; |
105 | __m128 dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22; |
106 | __m128 velec,felec,velecsum,facel,crf,krf,krf2; |
107 | real *charge; |
108 | int nvdwtype; |
109 | __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6; |
110 | int *vdwtype; |
111 | real *vdwparam; |
112 | __m128 one_sixth = _mm_set1_ps(1.0/6.0); |
113 | __m128 one_twelfth = _mm_set1_ps(1.0/12.0); |
114 | __m128i vfitab; |
115 | __m128i ifour = _mm_set1_epi32(4); |
116 | __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF; |
117 | real *vftab; |
118 | __m128 dummy_mask,cutoff_mask; |
119 | __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) ); |
120 | __m128 one = _mm_set1_ps(1.0); |
121 | __m128 two = _mm_set1_ps(2.0); |
122 | x = xx[0]; |
123 | f = ff[0]; |
124 | |
125 | nri = nlist->nri; |
126 | iinr = nlist->iinr; |
127 | jindex = nlist->jindex; |
128 | jjnr = nlist->jjnr; |
129 | shiftidx = nlist->shift; |
130 | gid = nlist->gid; |
131 | shiftvec = fr->shift_vec[0]; |
132 | fshift = fr->fshift[0]; |
133 | facel = _mm_set1_ps(fr->epsfac); |
134 | charge = mdatoms->chargeA; |
135 | krf = _mm_set1_ps(fr->ic->k_rf); |
136 | krf2 = _mm_set1_ps(fr->ic->k_rf*2.0); |
137 | crf = _mm_set1_ps(fr->ic->c_rf); |
138 | nvdwtype = fr->ntype; |
139 | vdwparam = fr->nbfp; |
140 | vdwtype = mdatoms->typeA; |
141 | |
142 | vftab = kernel_data->table_vdw->data; |
143 | vftabscale = _mm_set1_ps(kernel_data->table_vdw->scale); |
144 | |
145 | /* Setup water-specific parameters */ |
146 | inr = nlist->iinr[0]; |
147 | iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0])); |
148 | iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1])); |
149 | iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2])); |
150 | vdwioffset0 = 2*nvdwtype*vdwtype[inr+0]; |
151 | |
152 | jq0 = _mm_set1_ps(charge[inr+0]); |
153 | jq1 = _mm_set1_ps(charge[inr+1]); |
154 | jq2 = _mm_set1_ps(charge[inr+2]); |
155 | vdwjidx0A = 2*vdwtype[inr+0]; |
156 | qq00 = _mm_mul_ps(iq0,jq0); |
157 | c6_00 = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]); |
158 | c12_00 = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]); |
159 | qq01 = _mm_mul_ps(iq0,jq1); |
160 | qq02 = _mm_mul_ps(iq0,jq2); |
161 | qq10 = _mm_mul_ps(iq1,jq0); |
162 | qq11 = _mm_mul_ps(iq1,jq1); |
163 | qq12 = _mm_mul_ps(iq1,jq2); |
164 | qq20 = _mm_mul_ps(iq2,jq0); |
165 | qq21 = _mm_mul_ps(iq2,jq1); |
166 | qq22 = _mm_mul_ps(iq2,jq2); |
167 | |
168 | /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */ |
169 | rcutoff_scalar = fr->rcoulomb; |
170 | rcutoff = _mm_set1_ps(rcutoff_scalar); |
171 | rcutoff2 = _mm_mul_ps(rcutoff,rcutoff); |
172 | |
173 | /* Avoid stupid compiler warnings */ |
174 | jnrA = jnrB = jnrC = jnrD = 0; |
175 | j_coord_offsetA = 0; |
176 | j_coord_offsetB = 0; |
177 | j_coord_offsetC = 0; |
178 | j_coord_offsetD = 0; |
179 | |
180 | outeriter = 0; |
181 | inneriter = 0; |
182 | |
183 | for(iidx=0;iidx<4*DIM3;iidx++) |
184 | { |
185 | scratch[iidx] = 0.0; |
186 | } |
187 | |
188 | /* Start outer loop over neighborlists */ |
189 | for(iidx=0; iidx<nri; iidx++) |
190 | { |
191 | /* Load shift vector for this list */ |
192 | i_shift_offset = DIM3*shiftidx[iidx]; |
193 | |
194 | /* Load limits for loop over neighbors */ |
195 | j_index_start = jindex[iidx]; |
196 | j_index_end = jindex[iidx+1]; |
197 | |
198 | /* Get outer coordinate index */ |
199 | inr = iinr[iidx]; |
200 | i_coord_offset = DIM3*inr; |
201 | |
202 | /* Load i particle coords and add shift vector */ |
203 | gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset, |
204 | &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2); |
205 | |
206 | fix0 = _mm_setzero_ps(); |
207 | fiy0 = _mm_setzero_ps(); |
208 | fiz0 = _mm_setzero_ps(); |
209 | fix1 = _mm_setzero_ps(); |
210 | fiy1 = _mm_setzero_ps(); |
211 | fiz1 = _mm_setzero_ps(); |
212 | fix2 = _mm_setzero_ps(); |
213 | fiy2 = _mm_setzero_ps(); |
214 | fiz2 = _mm_setzero_ps(); |
215 | |
216 | /* Reset potential sums */ |
217 | velecsum = _mm_setzero_ps(); |
218 | vvdwsum = _mm_setzero_ps(); |
219 | |
220 | /* Start inner kernel loop */ |
221 | for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4) |
222 | { |
223 | |
224 | /* Get j neighbor index, and coordinate index */ |
225 | jnrA = jjnr[jidx]; |
226 | jnrB = jjnr[jidx+1]; |
227 | jnrC = jjnr[jidx+2]; |
228 | jnrD = jjnr[jidx+3]; |
229 | j_coord_offsetA = DIM3*jnrA; |
230 | j_coord_offsetB = DIM3*jnrB; |
231 | j_coord_offsetC = DIM3*jnrC; |
232 | j_coord_offsetD = DIM3*jnrD; |
233 | |
234 | /* load j atom coordinates */ |
235 | gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB, |
236 | x+j_coord_offsetC,x+j_coord_offsetD, |
237 | &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2); |
238 | |
239 | /* Calculate displacement vector */ |
240 | dx00 = _mm_sub_ps(ix0,jx0); |
241 | dy00 = _mm_sub_ps(iy0,jy0); |
242 | dz00 = _mm_sub_ps(iz0,jz0); |
243 | dx01 = _mm_sub_ps(ix0,jx1); |
244 | dy01 = _mm_sub_ps(iy0,jy1); |
245 | dz01 = _mm_sub_ps(iz0,jz1); |
246 | dx02 = _mm_sub_ps(ix0,jx2); |
247 | dy02 = _mm_sub_ps(iy0,jy2); |
248 | dz02 = _mm_sub_ps(iz0,jz2); |
249 | dx10 = _mm_sub_ps(ix1,jx0); |
250 | dy10 = _mm_sub_ps(iy1,jy0); |
251 | dz10 = _mm_sub_ps(iz1,jz0); |
252 | dx11 = _mm_sub_ps(ix1,jx1); |
253 | dy11 = _mm_sub_ps(iy1,jy1); |
254 | dz11 = _mm_sub_ps(iz1,jz1); |
255 | dx12 = _mm_sub_ps(ix1,jx2); |
256 | dy12 = _mm_sub_ps(iy1,jy2); |
257 | dz12 = _mm_sub_ps(iz1,jz2); |
258 | dx20 = _mm_sub_ps(ix2,jx0); |
259 | dy20 = _mm_sub_ps(iy2,jy0); |
260 | dz20 = _mm_sub_ps(iz2,jz0); |
261 | dx21 = _mm_sub_ps(ix2,jx1); |
262 | dy21 = _mm_sub_ps(iy2,jy1); |
263 | dz21 = _mm_sub_ps(iz2,jz1); |
264 | dx22 = _mm_sub_ps(ix2,jx2); |
265 | dy22 = _mm_sub_ps(iy2,jy2); |
266 | dz22 = _mm_sub_ps(iz2,jz2); |
267 | |
268 | /* Calculate squared distance and things based on it */ |
269 | rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00); |
270 | rsq01 = gmx_mm_calc_rsq_ps(dx01,dy01,dz01); |
271 | rsq02 = gmx_mm_calc_rsq_ps(dx02,dy02,dz02); |
272 | rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10); |
273 | rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11); |
274 | rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12); |
275 | rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20); |
276 | rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21); |
277 | rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22); |
278 | |
279 | rinv00 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq00); |
280 | rinv01 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq01); |
281 | rinv02 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq02); |
282 | rinv10 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq10); |
283 | rinv11 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq11); |
284 | rinv12 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq12); |
285 | rinv20 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq20); |
286 | rinv21 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq21); |
287 | rinv22 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq22); |
288 | |
289 | rinvsq00 = _mm_mul_ps(rinv00,rinv00); |
290 | rinvsq01 = _mm_mul_ps(rinv01,rinv01); |
291 | rinvsq02 = _mm_mul_ps(rinv02,rinv02); |
292 | rinvsq10 = _mm_mul_ps(rinv10,rinv10); |
293 | rinvsq11 = _mm_mul_ps(rinv11,rinv11); |
294 | rinvsq12 = _mm_mul_ps(rinv12,rinv12); |
295 | rinvsq20 = _mm_mul_ps(rinv20,rinv20); |
296 | rinvsq21 = _mm_mul_ps(rinv21,rinv21); |
297 | rinvsq22 = _mm_mul_ps(rinv22,rinv22); |
298 | |
299 | fjx0 = _mm_setzero_ps(); |
300 | fjy0 = _mm_setzero_ps(); |
301 | fjz0 = _mm_setzero_ps(); |
302 | fjx1 = _mm_setzero_ps(); |
303 | fjy1 = _mm_setzero_ps(); |
304 | fjz1 = _mm_setzero_ps(); |
305 | fjx2 = _mm_setzero_ps(); |
306 | fjy2 = _mm_setzero_ps(); |
307 | fjz2 = _mm_setzero_ps(); |
308 | |
309 | /************************** |
310 | * CALCULATE INTERACTIONS * |
311 | **************************/ |
312 | |
313 | if (gmx_mm_any_lt(rsq00,rcutoff2)) |
314 | { |
315 | |
316 | r00 = _mm_mul_ps(rsq00,rinv00); |
317 | |
318 | /* Calculate table index by multiplying r with table scale and truncate to integer */ |
319 | rt = _mm_mul_ps(r00,vftabscale); |
320 | vfitab = _mm_cvttps_epi32(rt); |
321 | vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (rt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
322 | vfitab = _mm_slli_epi32(vfitab,3); |
323 | |
324 | /* REACTION-FIELD ELECTROSTATICS */ |
325 | velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf)); |
326 | felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2)); |
327 | |
328 | /* CUBIC SPLINE TABLE DISPERSION */ |
329 | Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(0) & 3];})) ); |
330 | F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(1) & 3];})) ); |
331 | G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(2) & 3];})) ); |
332 | H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(3) & 3];})) ); |
333 | _MM_TRANSPOSE4_PS(Y,F,G,H)do { __m128 tmp3, tmp2, tmp1, tmp0; tmp0 = _mm_unpacklo_ps((Y ), (F)); tmp2 = _mm_unpacklo_ps((G), (H)); tmp1 = _mm_unpackhi_ps ((Y), (F)); tmp3 = _mm_unpackhi_ps((G), (H)); (Y) = _mm_movelh_ps (tmp0, tmp2); (F) = _mm_movehl_ps(tmp2, tmp0); (G) = _mm_movelh_ps (tmp1, tmp3); (H) = _mm_movehl_ps(tmp3, tmp1); } while (0); |
334 | Heps = _mm_mul_ps(vfeps,H); |
335 | Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps))); |
336 | VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp)); |
337 | vvdw6 = _mm_mul_ps(c6_00,VV); |
338 | FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps)))); |
339 | fvdw6 = _mm_mul_ps(c6_00,FF); |
340 | |
341 | /* CUBIC SPLINE TABLE REPULSION */ |
342 | vfitab = _mm_add_epi32(vfitab,ifour); |
343 | Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(0) & 3];})) ); |
344 | F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(1) & 3];})) ); |
345 | G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(2) & 3];})) ); |
346 | H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(3) & 3];})) ); |
347 | _MM_TRANSPOSE4_PS(Y,F,G,H)do { __m128 tmp3, tmp2, tmp1, tmp0; tmp0 = _mm_unpacklo_ps((Y ), (F)); tmp2 = _mm_unpacklo_ps((G), (H)); tmp1 = _mm_unpackhi_ps ((Y), (F)); tmp3 = _mm_unpackhi_ps((G), (H)); (Y) = _mm_movelh_ps (tmp0, tmp2); (F) = _mm_movehl_ps(tmp2, tmp0); (G) = _mm_movelh_ps (tmp1, tmp3); (H) = _mm_movehl_ps(tmp3, tmp1); } while (0); |
348 | Heps = _mm_mul_ps(vfeps,H); |
349 | Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps))); |
350 | VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp)); |
351 | vvdw12 = _mm_mul_ps(c12_00,VV); |
352 | FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps)))); |
353 | fvdw12 = _mm_mul_ps(c12_00,FF); |
354 | vvdw = _mm_add_ps(vvdw12,vvdw6); |
355 | fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00))); |
356 | |
357 | cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2); |
358 | |
359 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
360 | velec = _mm_and_ps(velec,cutoff_mask); |
361 | velecsum = _mm_add_ps(velecsum,velec); |
362 | vvdw = _mm_and_ps(vvdw,cutoff_mask); |
363 | vvdwsum = _mm_add_ps(vvdwsum,vvdw); |
364 | |
365 | fscal = _mm_add_ps(felec,fvdw); |
366 | |
367 | fscal = _mm_and_ps(fscal,cutoff_mask); |
368 | |
369 | /* Calculate temporary vectorial force */ |
370 | tx = _mm_mul_ps(fscal,dx00); |
371 | ty = _mm_mul_ps(fscal,dy00); |
372 | tz = _mm_mul_ps(fscal,dz00); |
373 | |
374 | /* Update vectorial force */ |
375 | fix0 = _mm_add_ps(fix0,tx); |
376 | fiy0 = _mm_add_ps(fiy0,ty); |
377 | fiz0 = _mm_add_ps(fiz0,tz); |
378 | |
379 | fjx0 = _mm_add_ps(fjx0,tx); |
380 | fjy0 = _mm_add_ps(fjy0,ty); |
381 | fjz0 = _mm_add_ps(fjz0,tz); |
382 | |
383 | } |
384 | |
385 | /************************** |
386 | * CALCULATE INTERACTIONS * |
387 | **************************/ |
388 | |
389 | if (gmx_mm_any_lt(rsq01,rcutoff2)) |
390 | { |
391 | |
392 | /* REACTION-FIELD ELECTROSTATICS */ |
393 | velec = _mm_mul_ps(qq01,_mm_sub_ps(_mm_add_ps(rinv01,_mm_mul_ps(krf,rsq01)),crf)); |
394 | felec = _mm_mul_ps(qq01,_mm_sub_ps(_mm_mul_ps(rinv01,rinvsq01),krf2)); |
395 | |
396 | cutoff_mask = _mm_cmplt_ps(rsq01,rcutoff2); |
397 | |
398 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
399 | velec = _mm_and_ps(velec,cutoff_mask); |
400 | velecsum = _mm_add_ps(velecsum,velec); |
401 | |
402 | fscal = felec; |
403 | |
404 | fscal = _mm_and_ps(fscal,cutoff_mask); |
405 | |
406 | /* Calculate temporary vectorial force */ |
407 | tx = _mm_mul_ps(fscal,dx01); |
408 | ty = _mm_mul_ps(fscal,dy01); |
409 | tz = _mm_mul_ps(fscal,dz01); |
410 | |
411 | /* Update vectorial force */ |
412 | fix0 = _mm_add_ps(fix0,tx); |
413 | fiy0 = _mm_add_ps(fiy0,ty); |
414 | fiz0 = _mm_add_ps(fiz0,tz); |
415 | |
416 | fjx1 = _mm_add_ps(fjx1,tx); |
417 | fjy1 = _mm_add_ps(fjy1,ty); |
418 | fjz1 = _mm_add_ps(fjz1,tz); |
419 | |
420 | } |
421 | |
422 | /************************** |
423 | * CALCULATE INTERACTIONS * |
424 | **************************/ |
425 | |
426 | if (gmx_mm_any_lt(rsq02,rcutoff2)) |
427 | { |
428 | |
429 | /* REACTION-FIELD ELECTROSTATICS */ |
430 | velec = _mm_mul_ps(qq02,_mm_sub_ps(_mm_add_ps(rinv02,_mm_mul_ps(krf,rsq02)),crf)); |
431 | felec = _mm_mul_ps(qq02,_mm_sub_ps(_mm_mul_ps(rinv02,rinvsq02),krf2)); |
432 | |
433 | cutoff_mask = _mm_cmplt_ps(rsq02,rcutoff2); |
434 | |
435 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
436 | velec = _mm_and_ps(velec,cutoff_mask); |
437 | velecsum = _mm_add_ps(velecsum,velec); |
438 | |
439 | fscal = felec; |
440 | |
441 | fscal = _mm_and_ps(fscal,cutoff_mask); |
442 | |
443 | /* Calculate temporary vectorial force */ |
444 | tx = _mm_mul_ps(fscal,dx02); |
445 | ty = _mm_mul_ps(fscal,dy02); |
446 | tz = _mm_mul_ps(fscal,dz02); |
447 | |
448 | /* Update vectorial force */ |
449 | fix0 = _mm_add_ps(fix0,tx); |
450 | fiy0 = _mm_add_ps(fiy0,ty); |
451 | fiz0 = _mm_add_ps(fiz0,tz); |
452 | |
453 | fjx2 = _mm_add_ps(fjx2,tx); |
454 | fjy2 = _mm_add_ps(fjy2,ty); |
455 | fjz2 = _mm_add_ps(fjz2,tz); |
456 | |
457 | } |
458 | |
459 | /************************** |
460 | * CALCULATE INTERACTIONS * |
461 | **************************/ |
462 | |
463 | if (gmx_mm_any_lt(rsq10,rcutoff2)) |
464 | { |
465 | |
466 | /* REACTION-FIELD ELECTROSTATICS */ |
467 | velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf)); |
468 | felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2)); |
469 | |
470 | cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2); |
471 | |
472 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
473 | velec = _mm_and_ps(velec,cutoff_mask); |
474 | velecsum = _mm_add_ps(velecsum,velec); |
475 | |
476 | fscal = felec; |
477 | |
478 | fscal = _mm_and_ps(fscal,cutoff_mask); |
479 | |
480 | /* Calculate temporary vectorial force */ |
481 | tx = _mm_mul_ps(fscal,dx10); |
482 | ty = _mm_mul_ps(fscal,dy10); |
483 | tz = _mm_mul_ps(fscal,dz10); |
484 | |
485 | /* Update vectorial force */ |
486 | fix1 = _mm_add_ps(fix1,tx); |
487 | fiy1 = _mm_add_ps(fiy1,ty); |
488 | fiz1 = _mm_add_ps(fiz1,tz); |
489 | |
490 | fjx0 = _mm_add_ps(fjx0,tx); |
491 | fjy0 = _mm_add_ps(fjy0,ty); |
492 | fjz0 = _mm_add_ps(fjz0,tz); |
493 | |
494 | } |
495 | |
496 | /************************** |
497 | * CALCULATE INTERACTIONS * |
498 | **************************/ |
499 | |
500 | if (gmx_mm_any_lt(rsq11,rcutoff2)) |
501 | { |
502 | |
503 | /* REACTION-FIELD ELECTROSTATICS */ |
504 | velec = _mm_mul_ps(qq11,_mm_sub_ps(_mm_add_ps(rinv11,_mm_mul_ps(krf,rsq11)),crf)); |
505 | felec = _mm_mul_ps(qq11,_mm_sub_ps(_mm_mul_ps(rinv11,rinvsq11),krf2)); |
506 | |
507 | cutoff_mask = _mm_cmplt_ps(rsq11,rcutoff2); |
508 | |
509 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
510 | velec = _mm_and_ps(velec,cutoff_mask); |
511 | velecsum = _mm_add_ps(velecsum,velec); |
512 | |
513 | fscal = felec; |
514 | |
515 | fscal = _mm_and_ps(fscal,cutoff_mask); |
516 | |
517 | /* Calculate temporary vectorial force */ |
518 | tx = _mm_mul_ps(fscal,dx11); |
519 | ty = _mm_mul_ps(fscal,dy11); |
520 | tz = _mm_mul_ps(fscal,dz11); |
521 | |
522 | /* Update vectorial force */ |
523 | fix1 = _mm_add_ps(fix1,tx); |
524 | fiy1 = _mm_add_ps(fiy1,ty); |
525 | fiz1 = _mm_add_ps(fiz1,tz); |
526 | |
527 | fjx1 = _mm_add_ps(fjx1,tx); |
528 | fjy1 = _mm_add_ps(fjy1,ty); |
529 | fjz1 = _mm_add_ps(fjz1,tz); |
530 | |
531 | } |
532 | |
533 | /************************** |
534 | * CALCULATE INTERACTIONS * |
535 | **************************/ |
536 | |
537 | if (gmx_mm_any_lt(rsq12,rcutoff2)) |
538 | { |
539 | |
540 | /* REACTION-FIELD ELECTROSTATICS */ |
541 | velec = _mm_mul_ps(qq12,_mm_sub_ps(_mm_add_ps(rinv12,_mm_mul_ps(krf,rsq12)),crf)); |
542 | felec = _mm_mul_ps(qq12,_mm_sub_ps(_mm_mul_ps(rinv12,rinvsq12),krf2)); |
543 | |
544 | cutoff_mask = _mm_cmplt_ps(rsq12,rcutoff2); |
545 | |
546 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
547 | velec = _mm_and_ps(velec,cutoff_mask); |
548 | velecsum = _mm_add_ps(velecsum,velec); |
549 | |
550 | fscal = felec; |
551 | |
552 | fscal = _mm_and_ps(fscal,cutoff_mask); |
553 | |
554 | /* Calculate temporary vectorial force */ |
555 | tx = _mm_mul_ps(fscal,dx12); |
556 | ty = _mm_mul_ps(fscal,dy12); |
557 | tz = _mm_mul_ps(fscal,dz12); |
558 | |
559 | /* Update vectorial force */ |
560 | fix1 = _mm_add_ps(fix1,tx); |
561 | fiy1 = _mm_add_ps(fiy1,ty); |
562 | fiz1 = _mm_add_ps(fiz1,tz); |
563 | |
564 | fjx2 = _mm_add_ps(fjx2,tx); |
565 | fjy2 = _mm_add_ps(fjy2,ty); |
566 | fjz2 = _mm_add_ps(fjz2,tz); |
567 | |
568 | } |
569 | |
570 | /************************** |
571 | * CALCULATE INTERACTIONS * |
572 | **************************/ |
573 | |
574 | if (gmx_mm_any_lt(rsq20,rcutoff2)) |
575 | { |
576 | |
577 | /* REACTION-FIELD ELECTROSTATICS */ |
578 | velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf)); |
579 | felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2)); |
580 | |
581 | cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2); |
582 | |
583 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
584 | velec = _mm_and_ps(velec,cutoff_mask); |
585 | velecsum = _mm_add_ps(velecsum,velec); |
586 | |
587 | fscal = felec; |
588 | |
589 | fscal = _mm_and_ps(fscal,cutoff_mask); |
590 | |
591 | /* Calculate temporary vectorial force */ |
592 | tx = _mm_mul_ps(fscal,dx20); |
593 | ty = _mm_mul_ps(fscal,dy20); |
594 | tz = _mm_mul_ps(fscal,dz20); |
595 | |
596 | /* Update vectorial force */ |
597 | fix2 = _mm_add_ps(fix2,tx); |
598 | fiy2 = _mm_add_ps(fiy2,ty); |
599 | fiz2 = _mm_add_ps(fiz2,tz); |
600 | |
601 | fjx0 = _mm_add_ps(fjx0,tx); |
602 | fjy0 = _mm_add_ps(fjy0,ty); |
603 | fjz0 = _mm_add_ps(fjz0,tz); |
604 | |
605 | } |
606 | |
607 | /************************** |
608 | * CALCULATE INTERACTIONS * |
609 | **************************/ |
610 | |
611 | if (gmx_mm_any_lt(rsq21,rcutoff2)) |
612 | { |
613 | |
614 | /* REACTION-FIELD ELECTROSTATICS */ |
615 | velec = _mm_mul_ps(qq21,_mm_sub_ps(_mm_add_ps(rinv21,_mm_mul_ps(krf,rsq21)),crf)); |
616 | felec = _mm_mul_ps(qq21,_mm_sub_ps(_mm_mul_ps(rinv21,rinvsq21),krf2)); |
617 | |
618 | cutoff_mask = _mm_cmplt_ps(rsq21,rcutoff2); |
619 | |
620 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
621 | velec = _mm_and_ps(velec,cutoff_mask); |
622 | velecsum = _mm_add_ps(velecsum,velec); |
623 | |
624 | fscal = felec; |
625 | |
626 | fscal = _mm_and_ps(fscal,cutoff_mask); |
627 | |
628 | /* Calculate temporary vectorial force */ |
629 | tx = _mm_mul_ps(fscal,dx21); |
630 | ty = _mm_mul_ps(fscal,dy21); |
631 | tz = _mm_mul_ps(fscal,dz21); |
632 | |
633 | /* Update vectorial force */ |
634 | fix2 = _mm_add_ps(fix2,tx); |
635 | fiy2 = _mm_add_ps(fiy2,ty); |
636 | fiz2 = _mm_add_ps(fiz2,tz); |
637 | |
638 | fjx1 = _mm_add_ps(fjx1,tx); |
639 | fjy1 = _mm_add_ps(fjy1,ty); |
640 | fjz1 = _mm_add_ps(fjz1,tz); |
641 | |
642 | } |
643 | |
644 | /************************** |
645 | * CALCULATE INTERACTIONS * |
646 | **************************/ |
647 | |
648 | if (gmx_mm_any_lt(rsq22,rcutoff2)) |
649 | { |
650 | |
651 | /* REACTION-FIELD ELECTROSTATICS */ |
652 | velec = _mm_mul_ps(qq22,_mm_sub_ps(_mm_add_ps(rinv22,_mm_mul_ps(krf,rsq22)),crf)); |
653 | felec = _mm_mul_ps(qq22,_mm_sub_ps(_mm_mul_ps(rinv22,rinvsq22),krf2)); |
654 | |
655 | cutoff_mask = _mm_cmplt_ps(rsq22,rcutoff2); |
656 | |
657 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
658 | velec = _mm_and_ps(velec,cutoff_mask); |
659 | velecsum = _mm_add_ps(velecsum,velec); |
660 | |
661 | fscal = felec; |
662 | |
663 | fscal = _mm_and_ps(fscal,cutoff_mask); |
664 | |
665 | /* Calculate temporary vectorial force */ |
666 | tx = _mm_mul_ps(fscal,dx22); |
667 | ty = _mm_mul_ps(fscal,dy22); |
668 | tz = _mm_mul_ps(fscal,dz22); |
669 | |
670 | /* Update vectorial force */ |
671 | fix2 = _mm_add_ps(fix2,tx); |
672 | fiy2 = _mm_add_ps(fiy2,ty); |
673 | fiz2 = _mm_add_ps(fiz2,tz); |
674 | |
675 | fjx2 = _mm_add_ps(fjx2,tx); |
676 | fjy2 = _mm_add_ps(fjy2,ty); |
677 | fjz2 = _mm_add_ps(fjz2,tz); |
678 | |
679 | } |
680 | |
681 | fjptrA = f+j_coord_offsetA; |
682 | fjptrB = f+j_coord_offsetB; |
683 | fjptrC = f+j_coord_offsetC; |
684 | fjptrD = f+j_coord_offsetD; |
685 | |
686 | gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD, |
687 | fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2); |
688 | |
689 | /* Inner loop uses 360 flops */ |
690 | } |
691 | |
692 | if(jidx<j_index_end) |
693 | { |
694 | |
695 | /* Get j neighbor index, and coordinate index */ |
696 | jnrlistA = jjnr[jidx]; |
697 | jnrlistB = jjnr[jidx+1]; |
698 | jnrlistC = jjnr[jidx+2]; |
699 | jnrlistD = jjnr[jidx+3]; |
700 | /* Sign of each element will be negative for non-real atoms. |
701 | * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones, |
702 | * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries. |
703 | */ |
704 | dummy_mask = gmx_mm_castsi128_ps_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())); |
705 | jnrA = (jnrlistA>=0) ? jnrlistA : 0; |
706 | jnrB = (jnrlistB>=0) ? jnrlistB : 0; |
707 | jnrC = (jnrlistC>=0) ? jnrlistC : 0; |
708 | jnrD = (jnrlistD>=0) ? jnrlistD : 0; |
709 | j_coord_offsetA = DIM3*jnrA; |
710 | j_coord_offsetB = DIM3*jnrB; |
711 | j_coord_offsetC = DIM3*jnrC; |
712 | j_coord_offsetD = DIM3*jnrD; |
713 | |
714 | /* load j atom coordinates */ |
715 | gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB, |
716 | x+j_coord_offsetC,x+j_coord_offsetD, |
717 | &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2); |
718 | |
719 | /* Calculate displacement vector */ |
720 | dx00 = _mm_sub_ps(ix0,jx0); |
721 | dy00 = _mm_sub_ps(iy0,jy0); |
722 | dz00 = _mm_sub_ps(iz0,jz0); |
723 | dx01 = _mm_sub_ps(ix0,jx1); |
724 | dy01 = _mm_sub_ps(iy0,jy1); |
725 | dz01 = _mm_sub_ps(iz0,jz1); |
726 | dx02 = _mm_sub_ps(ix0,jx2); |
727 | dy02 = _mm_sub_ps(iy0,jy2); |
728 | dz02 = _mm_sub_ps(iz0,jz2); |
729 | dx10 = _mm_sub_ps(ix1,jx0); |
730 | dy10 = _mm_sub_ps(iy1,jy0); |
731 | dz10 = _mm_sub_ps(iz1,jz0); |
732 | dx11 = _mm_sub_ps(ix1,jx1); |
733 | dy11 = _mm_sub_ps(iy1,jy1); |
734 | dz11 = _mm_sub_ps(iz1,jz1); |
735 | dx12 = _mm_sub_ps(ix1,jx2); |
736 | dy12 = _mm_sub_ps(iy1,jy2); |
737 | dz12 = _mm_sub_ps(iz1,jz2); |
738 | dx20 = _mm_sub_ps(ix2,jx0); |
739 | dy20 = _mm_sub_ps(iy2,jy0); |
740 | dz20 = _mm_sub_ps(iz2,jz0); |
741 | dx21 = _mm_sub_ps(ix2,jx1); |
742 | dy21 = _mm_sub_ps(iy2,jy1); |
743 | dz21 = _mm_sub_ps(iz2,jz1); |
744 | dx22 = _mm_sub_ps(ix2,jx2); |
745 | dy22 = _mm_sub_ps(iy2,jy2); |
746 | dz22 = _mm_sub_ps(iz2,jz2); |
747 | |
748 | /* Calculate squared distance and things based on it */ |
749 | rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00); |
750 | rsq01 = gmx_mm_calc_rsq_ps(dx01,dy01,dz01); |
751 | rsq02 = gmx_mm_calc_rsq_ps(dx02,dy02,dz02); |
752 | rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10); |
753 | rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11); |
754 | rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12); |
755 | rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20); |
756 | rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21); |
757 | rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22); |
758 | |
759 | rinv00 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq00); |
760 | rinv01 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq01); |
761 | rinv02 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq02); |
762 | rinv10 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq10); |
763 | rinv11 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq11); |
764 | rinv12 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq12); |
765 | rinv20 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq20); |
766 | rinv21 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq21); |
767 | rinv22 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq22); |
768 | |
769 | rinvsq00 = _mm_mul_ps(rinv00,rinv00); |
770 | rinvsq01 = _mm_mul_ps(rinv01,rinv01); |
771 | rinvsq02 = _mm_mul_ps(rinv02,rinv02); |
772 | rinvsq10 = _mm_mul_ps(rinv10,rinv10); |
773 | rinvsq11 = _mm_mul_ps(rinv11,rinv11); |
774 | rinvsq12 = _mm_mul_ps(rinv12,rinv12); |
775 | rinvsq20 = _mm_mul_ps(rinv20,rinv20); |
776 | rinvsq21 = _mm_mul_ps(rinv21,rinv21); |
777 | rinvsq22 = _mm_mul_ps(rinv22,rinv22); |
778 | |
779 | fjx0 = _mm_setzero_ps(); |
780 | fjy0 = _mm_setzero_ps(); |
781 | fjz0 = _mm_setzero_ps(); |
782 | fjx1 = _mm_setzero_ps(); |
783 | fjy1 = _mm_setzero_ps(); |
784 | fjz1 = _mm_setzero_ps(); |
785 | fjx2 = _mm_setzero_ps(); |
786 | fjy2 = _mm_setzero_ps(); |
787 | fjz2 = _mm_setzero_ps(); |
788 | |
789 | /************************** |
790 | * CALCULATE INTERACTIONS * |
791 | **************************/ |
792 | |
793 | if (gmx_mm_any_lt(rsq00,rcutoff2)) |
794 | { |
795 | |
796 | r00 = _mm_mul_ps(rsq00,rinv00); |
797 | r00 = _mm_andnot_ps(dummy_mask,r00); |
798 | |
799 | /* Calculate table index by multiplying r with table scale and truncate to integer */ |
800 | rt = _mm_mul_ps(r00,vftabscale); |
801 | vfitab = _mm_cvttps_epi32(rt); |
802 | vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (rt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
803 | vfitab = _mm_slli_epi32(vfitab,3); |
804 | |
805 | /* REACTION-FIELD ELECTROSTATICS */ |
806 | velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf)); |
807 | felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2)); |
808 | |
809 | /* CUBIC SPLINE TABLE DISPERSION */ |
810 | Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(0) & 3];})) ); |
811 | F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(1) & 3];})) ); |
812 | G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(2) & 3];})) ); |
813 | H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(3) & 3];})) ); |
814 | _MM_TRANSPOSE4_PS(Y,F,G,H)do { __m128 tmp3, tmp2, tmp1, tmp0; tmp0 = _mm_unpacklo_ps((Y ), (F)); tmp2 = _mm_unpacklo_ps((G), (H)); tmp1 = _mm_unpackhi_ps ((Y), (F)); tmp3 = _mm_unpackhi_ps((G), (H)); (Y) = _mm_movelh_ps (tmp0, tmp2); (F) = _mm_movehl_ps(tmp2, tmp0); (G) = _mm_movelh_ps (tmp1, tmp3); (H) = _mm_movehl_ps(tmp3, tmp1); } while (0); |
815 | Heps = _mm_mul_ps(vfeps,H); |
816 | Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps))); |
817 | VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp)); |
818 | vvdw6 = _mm_mul_ps(c6_00,VV); |
819 | FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps)))); |
820 | fvdw6 = _mm_mul_ps(c6_00,FF); |
821 | |
822 | /* CUBIC SPLINE TABLE REPULSION */ |
823 | vfitab = _mm_add_epi32(vfitab,ifour); |
824 | Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(0) & 3];})) ); |
825 | F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(1) & 3];})) ); |
826 | G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(2) & 3];})) ); |
827 | H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(3) & 3];})) ); |
828 | _MM_TRANSPOSE4_PS(Y,F,G,H)do { __m128 tmp3, tmp2, tmp1, tmp0; tmp0 = _mm_unpacklo_ps((Y ), (F)); tmp2 = _mm_unpacklo_ps((G), (H)); tmp1 = _mm_unpackhi_ps ((Y), (F)); tmp3 = _mm_unpackhi_ps((G), (H)); (Y) = _mm_movelh_ps (tmp0, tmp2); (F) = _mm_movehl_ps(tmp2, tmp0); (G) = _mm_movelh_ps (tmp1, tmp3); (H) = _mm_movehl_ps(tmp3, tmp1); } while (0); |
829 | Heps = _mm_mul_ps(vfeps,H); |
830 | Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps))); |
831 | VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp)); |
832 | vvdw12 = _mm_mul_ps(c12_00,VV); |
833 | FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps)))); |
834 | fvdw12 = _mm_mul_ps(c12_00,FF); |
835 | vvdw = _mm_add_ps(vvdw12,vvdw6); |
836 | fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00))); |
837 | |
838 | cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2); |
839 | |
840 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
841 | velec = _mm_and_ps(velec,cutoff_mask); |
842 | velec = _mm_andnot_ps(dummy_mask,velec); |
843 | velecsum = _mm_add_ps(velecsum,velec); |
844 | vvdw = _mm_and_ps(vvdw,cutoff_mask); |
845 | vvdw = _mm_andnot_ps(dummy_mask,vvdw); |
846 | vvdwsum = _mm_add_ps(vvdwsum,vvdw); |
847 | |
848 | fscal = _mm_add_ps(felec,fvdw); |
849 | |
850 | fscal = _mm_and_ps(fscal,cutoff_mask); |
851 | |
852 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
853 | |
854 | /* Calculate temporary vectorial force */ |
855 | tx = _mm_mul_ps(fscal,dx00); |
856 | ty = _mm_mul_ps(fscal,dy00); |
857 | tz = _mm_mul_ps(fscal,dz00); |
858 | |
859 | /* Update vectorial force */ |
860 | fix0 = _mm_add_ps(fix0,tx); |
861 | fiy0 = _mm_add_ps(fiy0,ty); |
862 | fiz0 = _mm_add_ps(fiz0,tz); |
863 | |
864 | fjx0 = _mm_add_ps(fjx0,tx); |
865 | fjy0 = _mm_add_ps(fjy0,ty); |
866 | fjz0 = _mm_add_ps(fjz0,tz); |
867 | |
868 | } |
869 | |
870 | /************************** |
871 | * CALCULATE INTERACTIONS * |
872 | **************************/ |
873 | |
874 | if (gmx_mm_any_lt(rsq01,rcutoff2)) |
875 | { |
876 | |
877 | /* REACTION-FIELD ELECTROSTATICS */ |
878 | velec = _mm_mul_ps(qq01,_mm_sub_ps(_mm_add_ps(rinv01,_mm_mul_ps(krf,rsq01)),crf)); |
879 | felec = _mm_mul_ps(qq01,_mm_sub_ps(_mm_mul_ps(rinv01,rinvsq01),krf2)); |
880 | |
881 | cutoff_mask = _mm_cmplt_ps(rsq01,rcutoff2); |
882 | |
883 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
884 | velec = _mm_and_ps(velec,cutoff_mask); |
885 | velec = _mm_andnot_ps(dummy_mask,velec); |
886 | velecsum = _mm_add_ps(velecsum,velec); |
887 | |
888 | fscal = felec; |
889 | |
890 | fscal = _mm_and_ps(fscal,cutoff_mask); |
891 | |
892 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
893 | |
894 | /* Calculate temporary vectorial force */ |
895 | tx = _mm_mul_ps(fscal,dx01); |
896 | ty = _mm_mul_ps(fscal,dy01); |
897 | tz = _mm_mul_ps(fscal,dz01); |
898 | |
899 | /* Update vectorial force */ |
900 | fix0 = _mm_add_ps(fix0,tx); |
901 | fiy0 = _mm_add_ps(fiy0,ty); |
902 | fiz0 = _mm_add_ps(fiz0,tz); |
903 | |
904 | fjx1 = _mm_add_ps(fjx1,tx); |
905 | fjy1 = _mm_add_ps(fjy1,ty); |
906 | fjz1 = _mm_add_ps(fjz1,tz); |
907 | |
908 | } |
909 | |
910 | /************************** |
911 | * CALCULATE INTERACTIONS * |
912 | **************************/ |
913 | |
914 | if (gmx_mm_any_lt(rsq02,rcutoff2)) |
915 | { |
916 | |
917 | /* REACTION-FIELD ELECTROSTATICS */ |
918 | velec = _mm_mul_ps(qq02,_mm_sub_ps(_mm_add_ps(rinv02,_mm_mul_ps(krf,rsq02)),crf)); |
919 | felec = _mm_mul_ps(qq02,_mm_sub_ps(_mm_mul_ps(rinv02,rinvsq02),krf2)); |
920 | |
921 | cutoff_mask = _mm_cmplt_ps(rsq02,rcutoff2); |
922 | |
923 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
924 | velec = _mm_and_ps(velec,cutoff_mask); |
925 | velec = _mm_andnot_ps(dummy_mask,velec); |
926 | velecsum = _mm_add_ps(velecsum,velec); |
927 | |
928 | fscal = felec; |
929 | |
930 | fscal = _mm_and_ps(fscal,cutoff_mask); |
931 | |
932 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
933 | |
934 | /* Calculate temporary vectorial force */ |
935 | tx = _mm_mul_ps(fscal,dx02); |
936 | ty = _mm_mul_ps(fscal,dy02); |
937 | tz = _mm_mul_ps(fscal,dz02); |
938 | |
939 | /* Update vectorial force */ |
940 | fix0 = _mm_add_ps(fix0,tx); |
941 | fiy0 = _mm_add_ps(fiy0,ty); |
942 | fiz0 = _mm_add_ps(fiz0,tz); |
943 | |
944 | fjx2 = _mm_add_ps(fjx2,tx); |
945 | fjy2 = _mm_add_ps(fjy2,ty); |
946 | fjz2 = _mm_add_ps(fjz2,tz); |
947 | |
948 | } |
949 | |
950 | /************************** |
951 | * CALCULATE INTERACTIONS * |
952 | **************************/ |
953 | |
954 | if (gmx_mm_any_lt(rsq10,rcutoff2)) |
955 | { |
956 | |
957 | /* REACTION-FIELD ELECTROSTATICS */ |
958 | velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf)); |
959 | felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2)); |
960 | |
961 | cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2); |
962 | |
963 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
964 | velec = _mm_and_ps(velec,cutoff_mask); |
965 | velec = _mm_andnot_ps(dummy_mask,velec); |
966 | velecsum = _mm_add_ps(velecsum,velec); |
967 | |
968 | fscal = felec; |
969 | |
970 | fscal = _mm_and_ps(fscal,cutoff_mask); |
971 | |
972 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
973 | |
974 | /* Calculate temporary vectorial force */ |
975 | tx = _mm_mul_ps(fscal,dx10); |
976 | ty = _mm_mul_ps(fscal,dy10); |
977 | tz = _mm_mul_ps(fscal,dz10); |
978 | |
979 | /* Update vectorial force */ |
980 | fix1 = _mm_add_ps(fix1,tx); |
981 | fiy1 = _mm_add_ps(fiy1,ty); |
982 | fiz1 = _mm_add_ps(fiz1,tz); |
983 | |
984 | fjx0 = _mm_add_ps(fjx0,tx); |
985 | fjy0 = _mm_add_ps(fjy0,ty); |
986 | fjz0 = _mm_add_ps(fjz0,tz); |
987 | |
988 | } |
989 | |
990 | /************************** |
991 | * CALCULATE INTERACTIONS * |
992 | **************************/ |
993 | |
994 | if (gmx_mm_any_lt(rsq11,rcutoff2)) |
995 | { |
996 | |
997 | /* REACTION-FIELD ELECTROSTATICS */ |
998 | velec = _mm_mul_ps(qq11,_mm_sub_ps(_mm_add_ps(rinv11,_mm_mul_ps(krf,rsq11)),crf)); |
999 | felec = _mm_mul_ps(qq11,_mm_sub_ps(_mm_mul_ps(rinv11,rinvsq11),krf2)); |
1000 | |
1001 | cutoff_mask = _mm_cmplt_ps(rsq11,rcutoff2); |
1002 | |
1003 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
1004 | velec = _mm_and_ps(velec,cutoff_mask); |
1005 | velec = _mm_andnot_ps(dummy_mask,velec); |
1006 | velecsum = _mm_add_ps(velecsum,velec); |
1007 | |
1008 | fscal = felec; |
1009 | |
1010 | fscal = _mm_and_ps(fscal,cutoff_mask); |
1011 | |
1012 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
1013 | |
1014 | /* Calculate temporary vectorial force */ |
1015 | tx = _mm_mul_ps(fscal,dx11); |
1016 | ty = _mm_mul_ps(fscal,dy11); |
1017 | tz = _mm_mul_ps(fscal,dz11); |
1018 | |
1019 | /* Update vectorial force */ |
1020 | fix1 = _mm_add_ps(fix1,tx); |
1021 | fiy1 = _mm_add_ps(fiy1,ty); |
1022 | fiz1 = _mm_add_ps(fiz1,tz); |
1023 | |
1024 | fjx1 = _mm_add_ps(fjx1,tx); |
1025 | fjy1 = _mm_add_ps(fjy1,ty); |
1026 | fjz1 = _mm_add_ps(fjz1,tz); |
1027 | |
1028 | } |
1029 | |
1030 | /************************** |
1031 | * CALCULATE INTERACTIONS * |
1032 | **************************/ |
1033 | |
1034 | if (gmx_mm_any_lt(rsq12,rcutoff2)) |
1035 | { |
1036 | |
1037 | /* REACTION-FIELD ELECTROSTATICS */ |
1038 | velec = _mm_mul_ps(qq12,_mm_sub_ps(_mm_add_ps(rinv12,_mm_mul_ps(krf,rsq12)),crf)); |
1039 | felec = _mm_mul_ps(qq12,_mm_sub_ps(_mm_mul_ps(rinv12,rinvsq12),krf2)); |
1040 | |
1041 | cutoff_mask = _mm_cmplt_ps(rsq12,rcutoff2); |
1042 | |
1043 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
1044 | velec = _mm_and_ps(velec,cutoff_mask); |
1045 | velec = _mm_andnot_ps(dummy_mask,velec); |
1046 | velecsum = _mm_add_ps(velecsum,velec); |
1047 | |
1048 | fscal = felec; |
1049 | |
1050 | fscal = _mm_and_ps(fscal,cutoff_mask); |
1051 | |
1052 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
1053 | |
1054 | /* Calculate temporary vectorial force */ |
1055 | tx = _mm_mul_ps(fscal,dx12); |
1056 | ty = _mm_mul_ps(fscal,dy12); |
1057 | tz = _mm_mul_ps(fscal,dz12); |
1058 | |
1059 | /* Update vectorial force */ |
1060 | fix1 = _mm_add_ps(fix1,tx); |
1061 | fiy1 = _mm_add_ps(fiy1,ty); |
1062 | fiz1 = _mm_add_ps(fiz1,tz); |
1063 | |
1064 | fjx2 = _mm_add_ps(fjx2,tx); |
1065 | fjy2 = _mm_add_ps(fjy2,ty); |
1066 | fjz2 = _mm_add_ps(fjz2,tz); |
1067 | |
1068 | } |
1069 | |
1070 | /************************** |
1071 | * CALCULATE INTERACTIONS * |
1072 | **************************/ |
1073 | |
1074 | if (gmx_mm_any_lt(rsq20,rcutoff2)) |
1075 | { |
1076 | |
1077 | /* REACTION-FIELD ELECTROSTATICS */ |
1078 | velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf)); |
1079 | felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2)); |
1080 | |
1081 | cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2); |
1082 | |
1083 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
1084 | velec = _mm_and_ps(velec,cutoff_mask); |
1085 | velec = _mm_andnot_ps(dummy_mask,velec); |
1086 | velecsum = _mm_add_ps(velecsum,velec); |
1087 | |
1088 | fscal = felec; |
1089 | |
1090 | fscal = _mm_and_ps(fscal,cutoff_mask); |
1091 | |
1092 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
1093 | |
1094 | /* Calculate temporary vectorial force */ |
1095 | tx = _mm_mul_ps(fscal,dx20); |
1096 | ty = _mm_mul_ps(fscal,dy20); |
1097 | tz = _mm_mul_ps(fscal,dz20); |
1098 | |
1099 | /* Update vectorial force */ |
1100 | fix2 = _mm_add_ps(fix2,tx); |
1101 | fiy2 = _mm_add_ps(fiy2,ty); |
1102 | fiz2 = _mm_add_ps(fiz2,tz); |
1103 | |
1104 | fjx0 = _mm_add_ps(fjx0,tx); |
1105 | fjy0 = _mm_add_ps(fjy0,ty); |
1106 | fjz0 = _mm_add_ps(fjz0,tz); |
1107 | |
1108 | } |
1109 | |
1110 | /************************** |
1111 | * CALCULATE INTERACTIONS * |
1112 | **************************/ |
1113 | |
1114 | if (gmx_mm_any_lt(rsq21,rcutoff2)) |
1115 | { |
1116 | |
1117 | /* REACTION-FIELD ELECTROSTATICS */ |
1118 | velec = _mm_mul_ps(qq21,_mm_sub_ps(_mm_add_ps(rinv21,_mm_mul_ps(krf,rsq21)),crf)); |
1119 | felec = _mm_mul_ps(qq21,_mm_sub_ps(_mm_mul_ps(rinv21,rinvsq21),krf2)); |
1120 | |
1121 | cutoff_mask = _mm_cmplt_ps(rsq21,rcutoff2); |
1122 | |
1123 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
1124 | velec = _mm_and_ps(velec,cutoff_mask); |
1125 | velec = _mm_andnot_ps(dummy_mask,velec); |
1126 | velecsum = _mm_add_ps(velecsum,velec); |
1127 | |
1128 | fscal = felec; |
1129 | |
1130 | fscal = _mm_and_ps(fscal,cutoff_mask); |
1131 | |
1132 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
1133 | |
1134 | /* Calculate temporary vectorial force */ |
1135 | tx = _mm_mul_ps(fscal,dx21); |
1136 | ty = _mm_mul_ps(fscal,dy21); |
1137 | tz = _mm_mul_ps(fscal,dz21); |
1138 | |
1139 | /* Update vectorial force */ |
1140 | fix2 = _mm_add_ps(fix2,tx); |
1141 | fiy2 = _mm_add_ps(fiy2,ty); |
1142 | fiz2 = _mm_add_ps(fiz2,tz); |
1143 | |
1144 | fjx1 = _mm_add_ps(fjx1,tx); |
1145 | fjy1 = _mm_add_ps(fjy1,ty); |
1146 | fjz1 = _mm_add_ps(fjz1,tz); |
1147 | |
1148 | } |
1149 | |
1150 | /************************** |
1151 | * CALCULATE INTERACTIONS * |
1152 | **************************/ |
1153 | |
1154 | if (gmx_mm_any_lt(rsq22,rcutoff2)) |
1155 | { |
1156 | |
1157 | /* REACTION-FIELD ELECTROSTATICS */ |
1158 | velec = _mm_mul_ps(qq22,_mm_sub_ps(_mm_add_ps(rinv22,_mm_mul_ps(krf,rsq22)),crf)); |
1159 | felec = _mm_mul_ps(qq22,_mm_sub_ps(_mm_mul_ps(rinv22,rinvsq22),krf2)); |
1160 | |
1161 | cutoff_mask = _mm_cmplt_ps(rsq22,rcutoff2); |
1162 | |
1163 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
1164 | velec = _mm_and_ps(velec,cutoff_mask); |
1165 | velec = _mm_andnot_ps(dummy_mask,velec); |
1166 | velecsum = _mm_add_ps(velecsum,velec); |
1167 | |
1168 | fscal = felec; |
1169 | |
1170 | fscal = _mm_and_ps(fscal,cutoff_mask); |
1171 | |
1172 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
1173 | |
1174 | /* Calculate temporary vectorial force */ |
1175 | tx = _mm_mul_ps(fscal,dx22); |
1176 | ty = _mm_mul_ps(fscal,dy22); |
1177 | tz = _mm_mul_ps(fscal,dz22); |
1178 | |
1179 | /* Update vectorial force */ |
1180 | fix2 = _mm_add_ps(fix2,tx); |
1181 | fiy2 = _mm_add_ps(fiy2,ty); |
1182 | fiz2 = _mm_add_ps(fiz2,tz); |
1183 | |
1184 | fjx2 = _mm_add_ps(fjx2,tx); |
1185 | fjy2 = _mm_add_ps(fjy2,ty); |
1186 | fjz2 = _mm_add_ps(fjz2,tz); |
1187 | |
1188 | } |
1189 | |
1190 | fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch; |
1191 | fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch; |
1192 | fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch; |
1193 | fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch; |
1194 | |
1195 | gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD, |
1196 | fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2); |
1197 | |
1198 | /* Inner loop uses 361 flops */ |
1199 | } |
1200 | |
1201 | /* End of innermost loop */ |
1202 | |
1203 | gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2, |
1204 | f+i_coord_offset,fshift+i_shift_offset); |
1205 | |
1206 | ggid = gid[iidx]; |
1207 | /* Update potential energies */ |
1208 | gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid); |
1209 | gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid); |
1210 | |
1211 | /* Increment number of inner iterations */ |
1212 | inneriter += j_index_end - j_index_start; |
1213 | |
1214 | /* Outer loop uses 20 flops */ |
1215 | } |
1216 | |
1217 | /* Increment number of outer iterations */ |
1218 | outeriter += nri; |
1219 | |
1220 | /* Update outer/inner flops */ |
1221 | |
1222 | inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*361)(nrnb)->n[eNR_NBKERNEL_ELEC_VDW_W3W3_VF] += outeriter*20 + inneriter*361; |
1223 | } |
1224 | /* |
1225 | * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwCSTab_GeomW3W3_F_sse4_1_single |
1226 | * Electrostatics interaction: ReactionField |
1227 | * VdW interaction: CubicSplineTable |
1228 | * Geometry: Water3-Water3 |
1229 | * Calculate force/pot: Force |
1230 | */ |
1231 | void |
1232 | nb_kernel_ElecRFCut_VdwCSTab_GeomW3W3_F_sse4_1_single |
1233 | (t_nblist * gmx_restrict nlist, |
1234 | rvec * gmx_restrict xx, |
1235 | rvec * gmx_restrict ff, |
1236 | t_forcerec * gmx_restrict fr, |
1237 | t_mdatoms * gmx_restrict mdatoms, |
1238 | nb_kernel_data_t gmx_unused__attribute__ ((unused)) * gmx_restrict kernel_data, |
1239 | t_nrnb * gmx_restrict nrnb) |
1240 | { |
1241 | /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or |
1242 | * just 0 for non-waters. |
1243 | * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different |
1244 | * jnr indices corresponding to data put in the four positions in the SIMD register. |
1245 | */ |
1246 | int i_shift_offset,i_coord_offset,outeriter,inneriter; |
1247 | int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx; |
1248 | int jnrA,jnrB,jnrC,jnrD; |
1249 | int jnrlistA,jnrlistB,jnrlistC,jnrlistD; |
1250 | int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD; |
1251 | int *iinr,*jindex,*jjnr,*shiftidx,*gid; |
1252 | real rcutoff_scalar; |
1253 | real *shiftvec,*fshift,*x,*f; |
1254 | real *fjptrA,*fjptrB,*fjptrC,*fjptrD; |
1255 | real scratch[4*DIM3]; |
1256 | __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall; |
1257 | int vdwioffset0; |
1258 | __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0; |
1259 | int vdwioffset1; |
1260 | __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1; |
1261 | int vdwioffset2; |
1262 | __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2; |
1263 | int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D; |
1264 | __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0; |
1265 | int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D; |
1266 | __m128 jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1; |
1267 | int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D; |
1268 | __m128 jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2; |
1269 | __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00; |
1270 | __m128 dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01; |
1271 | __m128 dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02; |
1272 | __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10; |
1273 | __m128 dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11; |
1274 | __m128 dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12; |
1275 | __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20; |
1276 | __m128 dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21; |
1277 | __m128 dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22; |
1278 | __m128 velec,felec,velecsum,facel,crf,krf,krf2; |
1279 | real *charge; |
1280 | int nvdwtype; |
1281 | __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6; |
1282 | int *vdwtype; |
1283 | real *vdwparam; |
1284 | __m128 one_sixth = _mm_set1_ps(1.0/6.0); |
Value stored to 'one_sixth' during its initialization is never read | |
1285 | __m128 one_twelfth = _mm_set1_ps(1.0/12.0); |
1286 | __m128i vfitab; |
1287 | __m128i ifour = _mm_set1_epi32(4); |
1288 | __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF; |
1289 | real *vftab; |
1290 | __m128 dummy_mask,cutoff_mask; |
1291 | __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) ); |
1292 | __m128 one = _mm_set1_ps(1.0); |
1293 | __m128 two = _mm_set1_ps(2.0); |
1294 | x = xx[0]; |
1295 | f = ff[0]; |
1296 | |
1297 | nri = nlist->nri; |
1298 | iinr = nlist->iinr; |
1299 | jindex = nlist->jindex; |
1300 | jjnr = nlist->jjnr; |
1301 | shiftidx = nlist->shift; |
1302 | gid = nlist->gid; |
1303 | shiftvec = fr->shift_vec[0]; |
1304 | fshift = fr->fshift[0]; |
1305 | facel = _mm_set1_ps(fr->epsfac); |
1306 | charge = mdatoms->chargeA; |
1307 | krf = _mm_set1_ps(fr->ic->k_rf); |
1308 | krf2 = _mm_set1_ps(fr->ic->k_rf*2.0); |
1309 | crf = _mm_set1_ps(fr->ic->c_rf); |
1310 | nvdwtype = fr->ntype; |
1311 | vdwparam = fr->nbfp; |
1312 | vdwtype = mdatoms->typeA; |
1313 | |
1314 | vftab = kernel_data->table_vdw->data; |
1315 | vftabscale = _mm_set1_ps(kernel_data->table_vdw->scale); |
1316 | |
1317 | /* Setup water-specific parameters */ |
1318 | inr = nlist->iinr[0]; |
1319 | iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0])); |
1320 | iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1])); |
1321 | iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2])); |
1322 | vdwioffset0 = 2*nvdwtype*vdwtype[inr+0]; |
1323 | |
1324 | jq0 = _mm_set1_ps(charge[inr+0]); |
1325 | jq1 = _mm_set1_ps(charge[inr+1]); |
1326 | jq2 = _mm_set1_ps(charge[inr+2]); |
1327 | vdwjidx0A = 2*vdwtype[inr+0]; |
1328 | qq00 = _mm_mul_ps(iq0,jq0); |
1329 | c6_00 = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]); |
1330 | c12_00 = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]); |
1331 | qq01 = _mm_mul_ps(iq0,jq1); |
1332 | qq02 = _mm_mul_ps(iq0,jq2); |
1333 | qq10 = _mm_mul_ps(iq1,jq0); |
1334 | qq11 = _mm_mul_ps(iq1,jq1); |
1335 | qq12 = _mm_mul_ps(iq1,jq2); |
1336 | qq20 = _mm_mul_ps(iq2,jq0); |
1337 | qq21 = _mm_mul_ps(iq2,jq1); |
1338 | qq22 = _mm_mul_ps(iq2,jq2); |
1339 | |
1340 | /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */ |
1341 | rcutoff_scalar = fr->rcoulomb; |
1342 | rcutoff = _mm_set1_ps(rcutoff_scalar); |
1343 | rcutoff2 = _mm_mul_ps(rcutoff,rcutoff); |
1344 | |
1345 | /* Avoid stupid compiler warnings */ |
1346 | jnrA = jnrB = jnrC = jnrD = 0; |
1347 | j_coord_offsetA = 0; |
1348 | j_coord_offsetB = 0; |
1349 | j_coord_offsetC = 0; |
1350 | j_coord_offsetD = 0; |
1351 | |
1352 | outeriter = 0; |
1353 | inneriter = 0; |
1354 | |
1355 | for(iidx=0;iidx<4*DIM3;iidx++) |
1356 | { |
1357 | scratch[iidx] = 0.0; |
1358 | } |
1359 | |
1360 | /* Start outer loop over neighborlists */ |
1361 | for(iidx=0; iidx<nri; iidx++) |
1362 | { |
1363 | /* Load shift vector for this list */ |
1364 | i_shift_offset = DIM3*shiftidx[iidx]; |
1365 | |
1366 | /* Load limits for loop over neighbors */ |
1367 | j_index_start = jindex[iidx]; |
1368 | j_index_end = jindex[iidx+1]; |
1369 | |
1370 | /* Get outer coordinate index */ |
1371 | inr = iinr[iidx]; |
1372 | i_coord_offset = DIM3*inr; |
1373 | |
1374 | /* Load i particle coords and add shift vector */ |
1375 | gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset, |
1376 | &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2); |
1377 | |
1378 | fix0 = _mm_setzero_ps(); |
1379 | fiy0 = _mm_setzero_ps(); |
1380 | fiz0 = _mm_setzero_ps(); |
1381 | fix1 = _mm_setzero_ps(); |
1382 | fiy1 = _mm_setzero_ps(); |
1383 | fiz1 = _mm_setzero_ps(); |
1384 | fix2 = _mm_setzero_ps(); |
1385 | fiy2 = _mm_setzero_ps(); |
1386 | fiz2 = _mm_setzero_ps(); |
1387 | |
1388 | /* Start inner kernel loop */ |
1389 | for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4) |
1390 | { |
1391 | |
1392 | /* Get j neighbor index, and coordinate index */ |
1393 | jnrA = jjnr[jidx]; |
1394 | jnrB = jjnr[jidx+1]; |
1395 | jnrC = jjnr[jidx+2]; |
1396 | jnrD = jjnr[jidx+3]; |
1397 | j_coord_offsetA = DIM3*jnrA; |
1398 | j_coord_offsetB = DIM3*jnrB; |
1399 | j_coord_offsetC = DIM3*jnrC; |
1400 | j_coord_offsetD = DIM3*jnrD; |
1401 | |
1402 | /* load j atom coordinates */ |
1403 | gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB, |
1404 | x+j_coord_offsetC,x+j_coord_offsetD, |
1405 | &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2); |
1406 | |
1407 | /* Calculate displacement vector */ |
1408 | dx00 = _mm_sub_ps(ix0,jx0); |
1409 | dy00 = _mm_sub_ps(iy0,jy0); |
1410 | dz00 = _mm_sub_ps(iz0,jz0); |
1411 | dx01 = _mm_sub_ps(ix0,jx1); |
1412 | dy01 = _mm_sub_ps(iy0,jy1); |
1413 | dz01 = _mm_sub_ps(iz0,jz1); |
1414 | dx02 = _mm_sub_ps(ix0,jx2); |
1415 | dy02 = _mm_sub_ps(iy0,jy2); |
1416 | dz02 = _mm_sub_ps(iz0,jz2); |
1417 | dx10 = _mm_sub_ps(ix1,jx0); |
1418 | dy10 = _mm_sub_ps(iy1,jy0); |
1419 | dz10 = _mm_sub_ps(iz1,jz0); |
1420 | dx11 = _mm_sub_ps(ix1,jx1); |
1421 | dy11 = _mm_sub_ps(iy1,jy1); |
1422 | dz11 = _mm_sub_ps(iz1,jz1); |
1423 | dx12 = _mm_sub_ps(ix1,jx2); |
1424 | dy12 = _mm_sub_ps(iy1,jy2); |
1425 | dz12 = _mm_sub_ps(iz1,jz2); |
1426 | dx20 = _mm_sub_ps(ix2,jx0); |
1427 | dy20 = _mm_sub_ps(iy2,jy0); |
1428 | dz20 = _mm_sub_ps(iz2,jz0); |
1429 | dx21 = _mm_sub_ps(ix2,jx1); |
1430 | dy21 = _mm_sub_ps(iy2,jy1); |
1431 | dz21 = _mm_sub_ps(iz2,jz1); |
1432 | dx22 = _mm_sub_ps(ix2,jx2); |
1433 | dy22 = _mm_sub_ps(iy2,jy2); |
1434 | dz22 = _mm_sub_ps(iz2,jz2); |
1435 | |
1436 | /* Calculate squared distance and things based on it */ |
1437 | rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00); |
1438 | rsq01 = gmx_mm_calc_rsq_ps(dx01,dy01,dz01); |
1439 | rsq02 = gmx_mm_calc_rsq_ps(dx02,dy02,dz02); |
1440 | rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10); |
1441 | rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11); |
1442 | rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12); |
1443 | rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20); |
1444 | rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21); |
1445 | rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22); |
1446 | |
1447 | rinv00 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq00); |
1448 | rinv01 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq01); |
1449 | rinv02 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq02); |
1450 | rinv10 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq10); |
1451 | rinv11 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq11); |
1452 | rinv12 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq12); |
1453 | rinv20 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq20); |
1454 | rinv21 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq21); |
1455 | rinv22 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq22); |
1456 | |
1457 | rinvsq00 = _mm_mul_ps(rinv00,rinv00); |
1458 | rinvsq01 = _mm_mul_ps(rinv01,rinv01); |
1459 | rinvsq02 = _mm_mul_ps(rinv02,rinv02); |
1460 | rinvsq10 = _mm_mul_ps(rinv10,rinv10); |
1461 | rinvsq11 = _mm_mul_ps(rinv11,rinv11); |
1462 | rinvsq12 = _mm_mul_ps(rinv12,rinv12); |
1463 | rinvsq20 = _mm_mul_ps(rinv20,rinv20); |
1464 | rinvsq21 = _mm_mul_ps(rinv21,rinv21); |
1465 | rinvsq22 = _mm_mul_ps(rinv22,rinv22); |
1466 | |
1467 | fjx0 = _mm_setzero_ps(); |
1468 | fjy0 = _mm_setzero_ps(); |
1469 | fjz0 = _mm_setzero_ps(); |
1470 | fjx1 = _mm_setzero_ps(); |
1471 | fjy1 = _mm_setzero_ps(); |
1472 | fjz1 = _mm_setzero_ps(); |
1473 | fjx2 = _mm_setzero_ps(); |
1474 | fjy2 = _mm_setzero_ps(); |
1475 | fjz2 = _mm_setzero_ps(); |
1476 | |
1477 | /************************** |
1478 | * CALCULATE INTERACTIONS * |
1479 | **************************/ |
1480 | |
1481 | if (gmx_mm_any_lt(rsq00,rcutoff2)) |
1482 | { |
1483 | |
1484 | r00 = _mm_mul_ps(rsq00,rinv00); |
1485 | |
1486 | /* Calculate table index by multiplying r with table scale and truncate to integer */ |
1487 | rt = _mm_mul_ps(r00,vftabscale); |
1488 | vfitab = _mm_cvttps_epi32(rt); |
1489 | vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (rt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
1490 | vfitab = _mm_slli_epi32(vfitab,3); |
1491 | |
1492 | /* REACTION-FIELD ELECTROSTATICS */ |
1493 | felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2)); |
1494 | |
1495 | /* CUBIC SPLINE TABLE DISPERSION */ |
1496 | Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(0) & 3];})) ); |
1497 | F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(1) & 3];})) ); |
1498 | G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(2) & 3];})) ); |
1499 | H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(3) & 3];})) ); |
1500 | _MM_TRANSPOSE4_PS(Y,F,G,H)do { __m128 tmp3, tmp2, tmp1, tmp0; tmp0 = _mm_unpacklo_ps((Y ), (F)); tmp2 = _mm_unpacklo_ps((G), (H)); tmp1 = _mm_unpackhi_ps ((Y), (F)); tmp3 = _mm_unpackhi_ps((G), (H)); (Y) = _mm_movelh_ps (tmp0, tmp2); (F) = _mm_movehl_ps(tmp2, tmp0); (G) = _mm_movelh_ps (tmp1, tmp3); (H) = _mm_movehl_ps(tmp3, tmp1); } while (0); |
1501 | Heps = _mm_mul_ps(vfeps,H); |
1502 | Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps))); |
1503 | FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps)))); |
1504 | fvdw6 = _mm_mul_ps(c6_00,FF); |
1505 | |
1506 | /* CUBIC SPLINE TABLE REPULSION */ |
1507 | vfitab = _mm_add_epi32(vfitab,ifour); |
1508 | Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(0) & 3];})) ); |
1509 | F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(1) & 3];})) ); |
1510 | G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(2) & 3];})) ); |
1511 | H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(3) & 3];})) ); |
1512 | _MM_TRANSPOSE4_PS(Y,F,G,H)do { __m128 tmp3, tmp2, tmp1, tmp0; tmp0 = _mm_unpacklo_ps((Y ), (F)); tmp2 = _mm_unpacklo_ps((G), (H)); tmp1 = _mm_unpackhi_ps ((Y), (F)); tmp3 = _mm_unpackhi_ps((G), (H)); (Y) = _mm_movelh_ps (tmp0, tmp2); (F) = _mm_movehl_ps(tmp2, tmp0); (G) = _mm_movelh_ps (tmp1, tmp3); (H) = _mm_movehl_ps(tmp3, tmp1); } while (0); |
1513 | Heps = _mm_mul_ps(vfeps,H); |
1514 | Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps))); |
1515 | FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps)))); |
1516 | fvdw12 = _mm_mul_ps(c12_00,FF); |
1517 | fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00))); |
1518 | |
1519 | cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2); |
1520 | |
1521 | fscal = _mm_add_ps(felec,fvdw); |
1522 | |
1523 | fscal = _mm_and_ps(fscal,cutoff_mask); |
1524 | |
1525 | /* Calculate temporary vectorial force */ |
1526 | tx = _mm_mul_ps(fscal,dx00); |
1527 | ty = _mm_mul_ps(fscal,dy00); |
1528 | tz = _mm_mul_ps(fscal,dz00); |
1529 | |
1530 | /* Update vectorial force */ |
1531 | fix0 = _mm_add_ps(fix0,tx); |
1532 | fiy0 = _mm_add_ps(fiy0,ty); |
1533 | fiz0 = _mm_add_ps(fiz0,tz); |
1534 | |
1535 | fjx0 = _mm_add_ps(fjx0,tx); |
1536 | fjy0 = _mm_add_ps(fjy0,ty); |
1537 | fjz0 = _mm_add_ps(fjz0,tz); |
1538 | |
1539 | } |
1540 | |
1541 | /************************** |
1542 | * CALCULATE INTERACTIONS * |
1543 | **************************/ |
1544 | |
1545 | if (gmx_mm_any_lt(rsq01,rcutoff2)) |
1546 | { |
1547 | |
1548 | /* REACTION-FIELD ELECTROSTATICS */ |
1549 | felec = _mm_mul_ps(qq01,_mm_sub_ps(_mm_mul_ps(rinv01,rinvsq01),krf2)); |
1550 | |
1551 | cutoff_mask = _mm_cmplt_ps(rsq01,rcutoff2); |
1552 | |
1553 | fscal = felec; |
1554 | |
1555 | fscal = _mm_and_ps(fscal,cutoff_mask); |
1556 | |
1557 | /* Calculate temporary vectorial force */ |
1558 | tx = _mm_mul_ps(fscal,dx01); |
1559 | ty = _mm_mul_ps(fscal,dy01); |
1560 | tz = _mm_mul_ps(fscal,dz01); |
1561 | |
1562 | /* Update vectorial force */ |
1563 | fix0 = _mm_add_ps(fix0,tx); |
1564 | fiy0 = _mm_add_ps(fiy0,ty); |
1565 | fiz0 = _mm_add_ps(fiz0,tz); |
1566 | |
1567 | fjx1 = _mm_add_ps(fjx1,tx); |
1568 | fjy1 = _mm_add_ps(fjy1,ty); |
1569 | fjz1 = _mm_add_ps(fjz1,tz); |
1570 | |
1571 | } |
1572 | |
1573 | /************************** |
1574 | * CALCULATE INTERACTIONS * |
1575 | **************************/ |
1576 | |
1577 | if (gmx_mm_any_lt(rsq02,rcutoff2)) |
1578 | { |
1579 | |
1580 | /* REACTION-FIELD ELECTROSTATICS */ |
1581 | felec = _mm_mul_ps(qq02,_mm_sub_ps(_mm_mul_ps(rinv02,rinvsq02),krf2)); |
1582 | |
1583 | cutoff_mask = _mm_cmplt_ps(rsq02,rcutoff2); |
1584 | |
1585 | fscal = felec; |
1586 | |
1587 | fscal = _mm_and_ps(fscal,cutoff_mask); |
1588 | |
1589 | /* Calculate temporary vectorial force */ |
1590 | tx = _mm_mul_ps(fscal,dx02); |
1591 | ty = _mm_mul_ps(fscal,dy02); |
1592 | tz = _mm_mul_ps(fscal,dz02); |
1593 | |
1594 | /* Update vectorial force */ |
1595 | fix0 = _mm_add_ps(fix0,tx); |
1596 | fiy0 = _mm_add_ps(fiy0,ty); |
1597 | fiz0 = _mm_add_ps(fiz0,tz); |
1598 | |
1599 | fjx2 = _mm_add_ps(fjx2,tx); |
1600 | fjy2 = _mm_add_ps(fjy2,ty); |
1601 | fjz2 = _mm_add_ps(fjz2,tz); |
1602 | |
1603 | } |
1604 | |
1605 | /************************** |
1606 | * CALCULATE INTERACTIONS * |
1607 | **************************/ |
1608 | |
1609 | if (gmx_mm_any_lt(rsq10,rcutoff2)) |
1610 | { |
1611 | |
1612 | /* REACTION-FIELD ELECTROSTATICS */ |
1613 | felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2)); |
1614 | |
1615 | cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2); |
1616 | |
1617 | fscal = felec; |
1618 | |
1619 | fscal = _mm_and_ps(fscal,cutoff_mask); |
1620 | |
1621 | /* Calculate temporary vectorial force */ |
1622 | tx = _mm_mul_ps(fscal,dx10); |
1623 | ty = _mm_mul_ps(fscal,dy10); |
1624 | tz = _mm_mul_ps(fscal,dz10); |
1625 | |
1626 | /* Update vectorial force */ |
1627 | fix1 = _mm_add_ps(fix1,tx); |
1628 | fiy1 = _mm_add_ps(fiy1,ty); |
1629 | fiz1 = _mm_add_ps(fiz1,tz); |
1630 | |
1631 | fjx0 = _mm_add_ps(fjx0,tx); |
1632 | fjy0 = _mm_add_ps(fjy0,ty); |
1633 | fjz0 = _mm_add_ps(fjz0,tz); |
1634 | |
1635 | } |
1636 | |
1637 | /************************** |
1638 | * CALCULATE INTERACTIONS * |
1639 | **************************/ |
1640 | |
1641 | if (gmx_mm_any_lt(rsq11,rcutoff2)) |
1642 | { |
1643 | |
1644 | /* REACTION-FIELD ELECTROSTATICS */ |
1645 | felec = _mm_mul_ps(qq11,_mm_sub_ps(_mm_mul_ps(rinv11,rinvsq11),krf2)); |
1646 | |
1647 | cutoff_mask = _mm_cmplt_ps(rsq11,rcutoff2); |
1648 | |
1649 | fscal = felec; |
1650 | |
1651 | fscal = _mm_and_ps(fscal,cutoff_mask); |
1652 | |
1653 | /* Calculate temporary vectorial force */ |
1654 | tx = _mm_mul_ps(fscal,dx11); |
1655 | ty = _mm_mul_ps(fscal,dy11); |
1656 | tz = _mm_mul_ps(fscal,dz11); |
1657 | |
1658 | /* Update vectorial force */ |
1659 | fix1 = _mm_add_ps(fix1,tx); |
1660 | fiy1 = _mm_add_ps(fiy1,ty); |
1661 | fiz1 = _mm_add_ps(fiz1,tz); |
1662 | |
1663 | fjx1 = _mm_add_ps(fjx1,tx); |
1664 | fjy1 = _mm_add_ps(fjy1,ty); |
1665 | fjz1 = _mm_add_ps(fjz1,tz); |
1666 | |
1667 | } |
1668 | |
1669 | /************************** |
1670 | * CALCULATE INTERACTIONS * |
1671 | **************************/ |
1672 | |
1673 | if (gmx_mm_any_lt(rsq12,rcutoff2)) |
1674 | { |
1675 | |
1676 | /* REACTION-FIELD ELECTROSTATICS */ |
1677 | felec = _mm_mul_ps(qq12,_mm_sub_ps(_mm_mul_ps(rinv12,rinvsq12),krf2)); |
1678 | |
1679 | cutoff_mask = _mm_cmplt_ps(rsq12,rcutoff2); |
1680 | |
1681 | fscal = felec; |
1682 | |
1683 | fscal = _mm_and_ps(fscal,cutoff_mask); |
1684 | |
1685 | /* Calculate temporary vectorial force */ |
1686 | tx = _mm_mul_ps(fscal,dx12); |
1687 | ty = _mm_mul_ps(fscal,dy12); |
1688 | tz = _mm_mul_ps(fscal,dz12); |
1689 | |
1690 | /* Update vectorial force */ |
1691 | fix1 = _mm_add_ps(fix1,tx); |
1692 | fiy1 = _mm_add_ps(fiy1,ty); |
1693 | fiz1 = _mm_add_ps(fiz1,tz); |
1694 | |
1695 | fjx2 = _mm_add_ps(fjx2,tx); |
1696 | fjy2 = _mm_add_ps(fjy2,ty); |
1697 | fjz2 = _mm_add_ps(fjz2,tz); |
1698 | |
1699 | } |
1700 | |
1701 | /************************** |
1702 | * CALCULATE INTERACTIONS * |
1703 | **************************/ |
1704 | |
1705 | if (gmx_mm_any_lt(rsq20,rcutoff2)) |
1706 | { |
1707 | |
1708 | /* REACTION-FIELD ELECTROSTATICS */ |
1709 | felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2)); |
1710 | |
1711 | cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2); |
1712 | |
1713 | fscal = felec; |
1714 | |
1715 | fscal = _mm_and_ps(fscal,cutoff_mask); |
1716 | |
1717 | /* Calculate temporary vectorial force */ |
1718 | tx = _mm_mul_ps(fscal,dx20); |
1719 | ty = _mm_mul_ps(fscal,dy20); |
1720 | tz = _mm_mul_ps(fscal,dz20); |
1721 | |
1722 | /* Update vectorial force */ |
1723 | fix2 = _mm_add_ps(fix2,tx); |
1724 | fiy2 = _mm_add_ps(fiy2,ty); |
1725 | fiz2 = _mm_add_ps(fiz2,tz); |
1726 | |
1727 | fjx0 = _mm_add_ps(fjx0,tx); |
1728 | fjy0 = _mm_add_ps(fjy0,ty); |
1729 | fjz0 = _mm_add_ps(fjz0,tz); |
1730 | |
1731 | } |
1732 | |
1733 | /************************** |
1734 | * CALCULATE INTERACTIONS * |
1735 | **************************/ |
1736 | |
1737 | if (gmx_mm_any_lt(rsq21,rcutoff2)) |
1738 | { |
1739 | |
1740 | /* REACTION-FIELD ELECTROSTATICS */ |
1741 | felec = _mm_mul_ps(qq21,_mm_sub_ps(_mm_mul_ps(rinv21,rinvsq21),krf2)); |
1742 | |
1743 | cutoff_mask = _mm_cmplt_ps(rsq21,rcutoff2); |
1744 | |
1745 | fscal = felec; |
1746 | |
1747 | fscal = _mm_and_ps(fscal,cutoff_mask); |
1748 | |
1749 | /* Calculate temporary vectorial force */ |
1750 | tx = _mm_mul_ps(fscal,dx21); |
1751 | ty = _mm_mul_ps(fscal,dy21); |
1752 | tz = _mm_mul_ps(fscal,dz21); |
1753 | |
1754 | /* Update vectorial force */ |
1755 | fix2 = _mm_add_ps(fix2,tx); |
1756 | fiy2 = _mm_add_ps(fiy2,ty); |
1757 | fiz2 = _mm_add_ps(fiz2,tz); |
1758 | |
1759 | fjx1 = _mm_add_ps(fjx1,tx); |
1760 | fjy1 = _mm_add_ps(fjy1,ty); |
1761 | fjz1 = _mm_add_ps(fjz1,tz); |
1762 | |
1763 | } |
1764 | |
1765 | /************************** |
1766 | * CALCULATE INTERACTIONS * |
1767 | **************************/ |
1768 | |
1769 | if (gmx_mm_any_lt(rsq22,rcutoff2)) |
1770 | { |
1771 | |
1772 | /* REACTION-FIELD ELECTROSTATICS */ |
1773 | felec = _mm_mul_ps(qq22,_mm_sub_ps(_mm_mul_ps(rinv22,rinvsq22),krf2)); |
1774 | |
1775 | cutoff_mask = _mm_cmplt_ps(rsq22,rcutoff2); |
1776 | |
1777 | fscal = felec; |
1778 | |
1779 | fscal = _mm_and_ps(fscal,cutoff_mask); |
1780 | |
1781 | /* Calculate temporary vectorial force */ |
1782 | tx = _mm_mul_ps(fscal,dx22); |
1783 | ty = _mm_mul_ps(fscal,dy22); |
1784 | tz = _mm_mul_ps(fscal,dz22); |
1785 | |
1786 | /* Update vectorial force */ |
1787 | fix2 = _mm_add_ps(fix2,tx); |
1788 | fiy2 = _mm_add_ps(fiy2,ty); |
1789 | fiz2 = _mm_add_ps(fiz2,tz); |
1790 | |
1791 | fjx2 = _mm_add_ps(fjx2,tx); |
1792 | fjy2 = _mm_add_ps(fjy2,ty); |
1793 | fjz2 = _mm_add_ps(fjz2,tz); |
1794 | |
1795 | } |
1796 | |
1797 | fjptrA = f+j_coord_offsetA; |
1798 | fjptrB = f+j_coord_offsetB; |
1799 | fjptrC = f+j_coord_offsetC; |
1800 | fjptrD = f+j_coord_offsetD; |
1801 | |
1802 | gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD, |
1803 | fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2); |
1804 | |
1805 | /* Inner loop uses 297 flops */ |
1806 | } |
1807 | |
1808 | if(jidx<j_index_end) |
1809 | { |
1810 | |
1811 | /* Get j neighbor index, and coordinate index */ |
1812 | jnrlistA = jjnr[jidx]; |
1813 | jnrlistB = jjnr[jidx+1]; |
1814 | jnrlistC = jjnr[jidx+2]; |
1815 | jnrlistD = jjnr[jidx+3]; |
1816 | /* Sign of each element will be negative for non-real atoms. |
1817 | * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones, |
1818 | * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries. |
1819 | */ |
1820 | dummy_mask = gmx_mm_castsi128_ps_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())); |
1821 | jnrA = (jnrlistA>=0) ? jnrlistA : 0; |
1822 | jnrB = (jnrlistB>=0) ? jnrlistB : 0; |
1823 | jnrC = (jnrlistC>=0) ? jnrlistC : 0; |
1824 | jnrD = (jnrlistD>=0) ? jnrlistD : 0; |
1825 | j_coord_offsetA = DIM3*jnrA; |
1826 | j_coord_offsetB = DIM3*jnrB; |
1827 | j_coord_offsetC = DIM3*jnrC; |
1828 | j_coord_offsetD = DIM3*jnrD; |
1829 | |
1830 | /* load j atom coordinates */ |
1831 | gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB, |
1832 | x+j_coord_offsetC,x+j_coord_offsetD, |
1833 | &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2); |
1834 | |
1835 | /* Calculate displacement vector */ |
1836 | dx00 = _mm_sub_ps(ix0,jx0); |
1837 | dy00 = _mm_sub_ps(iy0,jy0); |
1838 | dz00 = _mm_sub_ps(iz0,jz0); |
1839 | dx01 = _mm_sub_ps(ix0,jx1); |
1840 | dy01 = _mm_sub_ps(iy0,jy1); |
1841 | dz01 = _mm_sub_ps(iz0,jz1); |
1842 | dx02 = _mm_sub_ps(ix0,jx2); |
1843 | dy02 = _mm_sub_ps(iy0,jy2); |
1844 | dz02 = _mm_sub_ps(iz0,jz2); |
1845 | dx10 = _mm_sub_ps(ix1,jx0); |
1846 | dy10 = _mm_sub_ps(iy1,jy0); |
1847 | dz10 = _mm_sub_ps(iz1,jz0); |
1848 | dx11 = _mm_sub_ps(ix1,jx1); |
1849 | dy11 = _mm_sub_ps(iy1,jy1); |
1850 | dz11 = _mm_sub_ps(iz1,jz1); |
1851 | dx12 = _mm_sub_ps(ix1,jx2); |
1852 | dy12 = _mm_sub_ps(iy1,jy2); |
1853 | dz12 = _mm_sub_ps(iz1,jz2); |
1854 | dx20 = _mm_sub_ps(ix2,jx0); |
1855 | dy20 = _mm_sub_ps(iy2,jy0); |
1856 | dz20 = _mm_sub_ps(iz2,jz0); |
1857 | dx21 = _mm_sub_ps(ix2,jx1); |
1858 | dy21 = _mm_sub_ps(iy2,jy1); |
1859 | dz21 = _mm_sub_ps(iz2,jz1); |
1860 | dx22 = _mm_sub_ps(ix2,jx2); |
1861 | dy22 = _mm_sub_ps(iy2,jy2); |
1862 | dz22 = _mm_sub_ps(iz2,jz2); |
1863 | |
1864 | /* Calculate squared distance and things based on it */ |
1865 | rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00); |
1866 | rsq01 = gmx_mm_calc_rsq_ps(dx01,dy01,dz01); |
1867 | rsq02 = gmx_mm_calc_rsq_ps(dx02,dy02,dz02); |
1868 | rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10); |
1869 | rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11); |
1870 | rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12); |
1871 | rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20); |
1872 | rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21); |
1873 | rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22); |
1874 | |
1875 | rinv00 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq00); |
1876 | rinv01 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq01); |
1877 | rinv02 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq02); |
1878 | rinv10 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq10); |
1879 | rinv11 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq11); |
1880 | rinv12 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq12); |
1881 | rinv20 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq20); |
1882 | rinv21 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq21); |
1883 | rinv22 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq22); |
1884 | |
1885 | rinvsq00 = _mm_mul_ps(rinv00,rinv00); |
1886 | rinvsq01 = _mm_mul_ps(rinv01,rinv01); |
1887 | rinvsq02 = _mm_mul_ps(rinv02,rinv02); |
1888 | rinvsq10 = _mm_mul_ps(rinv10,rinv10); |
1889 | rinvsq11 = _mm_mul_ps(rinv11,rinv11); |
1890 | rinvsq12 = _mm_mul_ps(rinv12,rinv12); |
1891 | rinvsq20 = _mm_mul_ps(rinv20,rinv20); |
1892 | rinvsq21 = _mm_mul_ps(rinv21,rinv21); |
1893 | rinvsq22 = _mm_mul_ps(rinv22,rinv22); |
1894 | |
1895 | fjx0 = _mm_setzero_ps(); |
1896 | fjy0 = _mm_setzero_ps(); |
1897 | fjz0 = _mm_setzero_ps(); |
1898 | fjx1 = _mm_setzero_ps(); |
1899 | fjy1 = _mm_setzero_ps(); |
1900 | fjz1 = _mm_setzero_ps(); |
1901 | fjx2 = _mm_setzero_ps(); |
1902 | fjy2 = _mm_setzero_ps(); |
1903 | fjz2 = _mm_setzero_ps(); |
1904 | |
1905 | /************************** |
1906 | * CALCULATE INTERACTIONS * |
1907 | **************************/ |
1908 | |
1909 | if (gmx_mm_any_lt(rsq00,rcutoff2)) |
1910 | { |
1911 | |
1912 | r00 = _mm_mul_ps(rsq00,rinv00); |
1913 | r00 = _mm_andnot_ps(dummy_mask,r00); |
1914 | |
1915 | /* Calculate table index by multiplying r with table scale and truncate to integer */ |
1916 | rt = _mm_mul_ps(r00,vftabscale); |
1917 | vfitab = _mm_cvttps_epi32(rt); |
1918 | vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (rt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
1919 | vfitab = _mm_slli_epi32(vfitab,3); |
1920 | |
1921 | /* REACTION-FIELD ELECTROSTATICS */ |
1922 | felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2)); |
1923 | |
1924 | /* CUBIC SPLINE TABLE DISPERSION */ |
1925 | Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(0) & 3];})) ); |
1926 | F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(1) & 3];})) ); |
1927 | G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(2) & 3];})) ); |
1928 | H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(3) & 3];})) ); |
1929 | _MM_TRANSPOSE4_PS(Y,F,G,H)do { __m128 tmp3, tmp2, tmp1, tmp0; tmp0 = _mm_unpacklo_ps((Y ), (F)); tmp2 = _mm_unpacklo_ps((G), (H)); tmp1 = _mm_unpackhi_ps ((Y), (F)); tmp3 = _mm_unpackhi_ps((G), (H)); (Y) = _mm_movelh_ps (tmp0, tmp2); (F) = _mm_movehl_ps(tmp2, tmp0); (G) = _mm_movelh_ps (tmp1, tmp3); (H) = _mm_movehl_ps(tmp3, tmp1); } while (0); |
1930 | Heps = _mm_mul_ps(vfeps,H); |
1931 | Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps))); |
1932 | FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps)))); |
1933 | fvdw6 = _mm_mul_ps(c6_00,FF); |
1934 | |
1935 | /* CUBIC SPLINE TABLE REPULSION */ |
1936 | vfitab = _mm_add_epi32(vfitab,ifour); |
1937 | Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(0) & 3];})) ); |
1938 | F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(1) & 3];})) ); |
1939 | G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(2) & 3];})) ); |
1940 | H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(3) & 3];})) ); |
1941 | _MM_TRANSPOSE4_PS(Y,F,G,H)do { __m128 tmp3, tmp2, tmp1, tmp0; tmp0 = _mm_unpacklo_ps((Y ), (F)); tmp2 = _mm_unpacklo_ps((G), (H)); tmp1 = _mm_unpackhi_ps ((Y), (F)); tmp3 = _mm_unpackhi_ps((G), (H)); (Y) = _mm_movelh_ps (tmp0, tmp2); (F) = _mm_movehl_ps(tmp2, tmp0); (G) = _mm_movelh_ps (tmp1, tmp3); (H) = _mm_movehl_ps(tmp3, tmp1); } while (0); |
1942 | Heps = _mm_mul_ps(vfeps,H); |
1943 | Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps))); |
1944 | FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps)))); |
1945 | fvdw12 = _mm_mul_ps(c12_00,FF); |
1946 | fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00))); |
1947 | |
1948 | cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2); |
1949 | |
1950 | fscal = _mm_add_ps(felec,fvdw); |
1951 | |
1952 | fscal = _mm_and_ps(fscal,cutoff_mask); |
1953 | |
1954 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
1955 | |
1956 | /* Calculate temporary vectorial force */ |
1957 | tx = _mm_mul_ps(fscal,dx00); |
1958 | ty = _mm_mul_ps(fscal,dy00); |
1959 | tz = _mm_mul_ps(fscal,dz00); |
1960 | |
1961 | /* Update vectorial force */ |
1962 | fix0 = _mm_add_ps(fix0,tx); |
1963 | fiy0 = _mm_add_ps(fiy0,ty); |
1964 | fiz0 = _mm_add_ps(fiz0,tz); |
1965 | |
1966 | fjx0 = _mm_add_ps(fjx0,tx); |
1967 | fjy0 = _mm_add_ps(fjy0,ty); |
1968 | fjz0 = _mm_add_ps(fjz0,tz); |
1969 | |
1970 | } |
1971 | |
1972 | /************************** |
1973 | * CALCULATE INTERACTIONS * |
1974 | **************************/ |
1975 | |
1976 | if (gmx_mm_any_lt(rsq01,rcutoff2)) |
1977 | { |
1978 | |
1979 | /* REACTION-FIELD ELECTROSTATICS */ |
1980 | felec = _mm_mul_ps(qq01,_mm_sub_ps(_mm_mul_ps(rinv01,rinvsq01),krf2)); |
1981 | |
1982 | cutoff_mask = _mm_cmplt_ps(rsq01,rcutoff2); |
1983 | |
1984 | fscal = felec; |
1985 | |
1986 | fscal = _mm_and_ps(fscal,cutoff_mask); |
1987 | |
1988 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
1989 | |
1990 | /* Calculate temporary vectorial force */ |
1991 | tx = _mm_mul_ps(fscal,dx01); |
1992 | ty = _mm_mul_ps(fscal,dy01); |
1993 | tz = _mm_mul_ps(fscal,dz01); |
1994 | |
1995 | /* Update vectorial force */ |
1996 | fix0 = _mm_add_ps(fix0,tx); |
1997 | fiy0 = _mm_add_ps(fiy0,ty); |
1998 | fiz0 = _mm_add_ps(fiz0,tz); |
1999 | |
2000 | fjx1 = _mm_add_ps(fjx1,tx); |
2001 | fjy1 = _mm_add_ps(fjy1,ty); |
2002 | fjz1 = _mm_add_ps(fjz1,tz); |
2003 | |
2004 | } |
2005 | |
2006 | /************************** |
2007 | * CALCULATE INTERACTIONS * |
2008 | **************************/ |
2009 | |
2010 | if (gmx_mm_any_lt(rsq02,rcutoff2)) |
2011 | { |
2012 | |
2013 | /* REACTION-FIELD ELECTROSTATICS */ |
2014 | felec = _mm_mul_ps(qq02,_mm_sub_ps(_mm_mul_ps(rinv02,rinvsq02),krf2)); |
2015 | |
2016 | cutoff_mask = _mm_cmplt_ps(rsq02,rcutoff2); |
2017 | |
2018 | fscal = felec; |
2019 | |
2020 | fscal = _mm_and_ps(fscal,cutoff_mask); |
2021 | |
2022 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
2023 | |
2024 | /* Calculate temporary vectorial force */ |
2025 | tx = _mm_mul_ps(fscal,dx02); |
2026 | ty = _mm_mul_ps(fscal,dy02); |
2027 | tz = _mm_mul_ps(fscal,dz02); |
2028 | |
2029 | /* Update vectorial force */ |
2030 | fix0 = _mm_add_ps(fix0,tx); |
2031 | fiy0 = _mm_add_ps(fiy0,ty); |
2032 | fiz0 = _mm_add_ps(fiz0,tz); |
2033 | |
2034 | fjx2 = _mm_add_ps(fjx2,tx); |
2035 | fjy2 = _mm_add_ps(fjy2,ty); |
2036 | fjz2 = _mm_add_ps(fjz2,tz); |
2037 | |
2038 | } |
2039 | |
2040 | /************************** |
2041 | * CALCULATE INTERACTIONS * |
2042 | **************************/ |
2043 | |
2044 | if (gmx_mm_any_lt(rsq10,rcutoff2)) |
2045 | { |
2046 | |
2047 | /* REACTION-FIELD ELECTROSTATICS */ |
2048 | felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2)); |
2049 | |
2050 | cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2); |
2051 | |
2052 | fscal = felec; |
2053 | |
2054 | fscal = _mm_and_ps(fscal,cutoff_mask); |
2055 | |
2056 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
2057 | |
2058 | /* Calculate temporary vectorial force */ |
2059 | tx = _mm_mul_ps(fscal,dx10); |
2060 | ty = _mm_mul_ps(fscal,dy10); |
2061 | tz = _mm_mul_ps(fscal,dz10); |
2062 | |
2063 | /* Update vectorial force */ |
2064 | fix1 = _mm_add_ps(fix1,tx); |
2065 | fiy1 = _mm_add_ps(fiy1,ty); |
2066 | fiz1 = _mm_add_ps(fiz1,tz); |
2067 | |
2068 | fjx0 = _mm_add_ps(fjx0,tx); |
2069 | fjy0 = _mm_add_ps(fjy0,ty); |
2070 | fjz0 = _mm_add_ps(fjz0,tz); |
2071 | |
2072 | } |
2073 | |
2074 | /************************** |
2075 | * CALCULATE INTERACTIONS * |
2076 | **************************/ |
2077 | |
2078 | if (gmx_mm_any_lt(rsq11,rcutoff2)) |
2079 | { |
2080 | |
2081 | /* REACTION-FIELD ELECTROSTATICS */ |
2082 | felec = _mm_mul_ps(qq11,_mm_sub_ps(_mm_mul_ps(rinv11,rinvsq11),krf2)); |
2083 | |
2084 | cutoff_mask = _mm_cmplt_ps(rsq11,rcutoff2); |
2085 | |
2086 | fscal = felec; |
2087 | |
2088 | fscal = _mm_and_ps(fscal,cutoff_mask); |
2089 | |
2090 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
2091 | |
2092 | /* Calculate temporary vectorial force */ |
2093 | tx = _mm_mul_ps(fscal,dx11); |
2094 | ty = _mm_mul_ps(fscal,dy11); |
2095 | tz = _mm_mul_ps(fscal,dz11); |
2096 | |
2097 | /* Update vectorial force */ |
2098 | fix1 = _mm_add_ps(fix1,tx); |
2099 | fiy1 = _mm_add_ps(fiy1,ty); |
2100 | fiz1 = _mm_add_ps(fiz1,tz); |
2101 | |
2102 | fjx1 = _mm_add_ps(fjx1,tx); |
2103 | fjy1 = _mm_add_ps(fjy1,ty); |
2104 | fjz1 = _mm_add_ps(fjz1,tz); |
2105 | |
2106 | } |
2107 | |
2108 | /************************** |
2109 | * CALCULATE INTERACTIONS * |
2110 | **************************/ |
2111 | |
2112 | if (gmx_mm_any_lt(rsq12,rcutoff2)) |
2113 | { |
2114 | |
2115 | /* REACTION-FIELD ELECTROSTATICS */ |
2116 | felec = _mm_mul_ps(qq12,_mm_sub_ps(_mm_mul_ps(rinv12,rinvsq12),krf2)); |
2117 | |
2118 | cutoff_mask = _mm_cmplt_ps(rsq12,rcutoff2); |
2119 | |
2120 | fscal = felec; |
2121 | |
2122 | fscal = _mm_and_ps(fscal,cutoff_mask); |
2123 | |
2124 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
2125 | |
2126 | /* Calculate temporary vectorial force */ |
2127 | tx = _mm_mul_ps(fscal,dx12); |
2128 | ty = _mm_mul_ps(fscal,dy12); |
2129 | tz = _mm_mul_ps(fscal,dz12); |
2130 | |
2131 | /* Update vectorial force */ |
2132 | fix1 = _mm_add_ps(fix1,tx); |
2133 | fiy1 = _mm_add_ps(fiy1,ty); |
2134 | fiz1 = _mm_add_ps(fiz1,tz); |
2135 | |
2136 | fjx2 = _mm_add_ps(fjx2,tx); |
2137 | fjy2 = _mm_add_ps(fjy2,ty); |
2138 | fjz2 = _mm_add_ps(fjz2,tz); |
2139 | |
2140 | } |
2141 | |
2142 | /************************** |
2143 | * CALCULATE INTERACTIONS * |
2144 | **************************/ |
2145 | |
2146 | if (gmx_mm_any_lt(rsq20,rcutoff2)) |
2147 | { |
2148 | |
2149 | /* REACTION-FIELD ELECTROSTATICS */ |
2150 | felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2)); |
2151 | |
2152 | cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2); |
2153 | |
2154 | fscal = felec; |
2155 | |
2156 | fscal = _mm_and_ps(fscal,cutoff_mask); |
2157 | |
2158 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
2159 | |
2160 | /* Calculate temporary vectorial force */ |
2161 | tx = _mm_mul_ps(fscal,dx20); |
2162 | ty = _mm_mul_ps(fscal,dy20); |
2163 | tz = _mm_mul_ps(fscal,dz20); |
2164 | |
2165 | /* Update vectorial force */ |
2166 | fix2 = _mm_add_ps(fix2,tx); |
2167 | fiy2 = _mm_add_ps(fiy2,ty); |
2168 | fiz2 = _mm_add_ps(fiz2,tz); |
2169 | |
2170 | fjx0 = _mm_add_ps(fjx0,tx); |
2171 | fjy0 = _mm_add_ps(fjy0,ty); |
2172 | fjz0 = _mm_add_ps(fjz0,tz); |
2173 | |
2174 | } |
2175 | |
2176 | /************************** |
2177 | * CALCULATE INTERACTIONS * |
2178 | **************************/ |
2179 | |
2180 | if (gmx_mm_any_lt(rsq21,rcutoff2)) |
2181 | { |
2182 | |
2183 | /* REACTION-FIELD ELECTROSTATICS */ |
2184 | felec = _mm_mul_ps(qq21,_mm_sub_ps(_mm_mul_ps(rinv21,rinvsq21),krf2)); |
2185 | |
2186 | cutoff_mask = _mm_cmplt_ps(rsq21,rcutoff2); |
2187 | |
2188 | fscal = felec; |
2189 | |
2190 | fscal = _mm_and_ps(fscal,cutoff_mask); |
2191 | |
2192 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
2193 | |
2194 | /* Calculate temporary vectorial force */ |
2195 | tx = _mm_mul_ps(fscal,dx21); |
2196 | ty = _mm_mul_ps(fscal,dy21); |
2197 | tz = _mm_mul_ps(fscal,dz21); |
2198 | |
2199 | /* Update vectorial force */ |
2200 | fix2 = _mm_add_ps(fix2,tx); |
2201 | fiy2 = _mm_add_ps(fiy2,ty); |
2202 | fiz2 = _mm_add_ps(fiz2,tz); |
2203 | |
2204 | fjx1 = _mm_add_ps(fjx1,tx); |
2205 | fjy1 = _mm_add_ps(fjy1,ty); |
2206 | fjz1 = _mm_add_ps(fjz1,tz); |
2207 | |
2208 | } |
2209 | |
2210 | /************************** |
2211 | * CALCULATE INTERACTIONS * |
2212 | **************************/ |
2213 | |
2214 | if (gmx_mm_any_lt(rsq22,rcutoff2)) |
2215 | { |
2216 | |
2217 | /* REACTION-FIELD ELECTROSTATICS */ |
2218 | felec = _mm_mul_ps(qq22,_mm_sub_ps(_mm_mul_ps(rinv22,rinvsq22),krf2)); |
2219 | |
2220 | cutoff_mask = _mm_cmplt_ps(rsq22,rcutoff2); |
2221 | |
2222 | fscal = felec; |
2223 | |
2224 | fscal = _mm_and_ps(fscal,cutoff_mask); |
2225 | |
2226 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
2227 | |
2228 | /* Calculate temporary vectorial force */ |
2229 | tx = _mm_mul_ps(fscal,dx22); |
2230 | ty = _mm_mul_ps(fscal,dy22); |
2231 | tz = _mm_mul_ps(fscal,dz22); |
2232 | |
2233 | /* Update vectorial force */ |
2234 | fix2 = _mm_add_ps(fix2,tx); |
2235 | fiy2 = _mm_add_ps(fiy2,ty); |
2236 | fiz2 = _mm_add_ps(fiz2,tz); |
2237 | |
2238 | fjx2 = _mm_add_ps(fjx2,tx); |
2239 | fjy2 = _mm_add_ps(fjy2,ty); |
2240 | fjz2 = _mm_add_ps(fjz2,tz); |
2241 | |
2242 | } |
2243 | |
2244 | fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch; |
2245 | fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch; |
2246 | fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch; |
2247 | fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch; |
2248 | |
2249 | gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD, |
2250 | fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2); |
2251 | |
2252 | /* Inner loop uses 298 flops */ |
2253 | } |
2254 | |
2255 | /* End of innermost loop */ |
2256 | |
2257 | gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2, |
2258 | f+i_coord_offset,fshift+i_shift_offset); |
2259 | |
2260 | /* Increment number of inner iterations */ |
2261 | inneriter += j_index_end - j_index_start; |
2262 | |
2263 | /* Outer loop uses 18 flops */ |
2264 | } |
2265 | |
2266 | /* Increment number of outer iterations */ |
2267 | outeriter += nri; |
2268 | |
2269 | /* Update outer/inner flops */ |
2270 | |
2271 | inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*298)(nrnb)->n[eNR_NBKERNEL_ELEC_VDW_W3W3_F] += outeriter*18 + inneriter *298; |
2272 | } |