1 | /* |
2 | * SparseGrid_chunk_copy.hpp |
3 | * |
4 | * Created on: Mar 18, 2020 |
5 | * Author: i-bird |
6 | */ |
7 | |
8 | #ifndef SPARSEGRID_CHUNK_COPY_HPP_ |
9 | #define SPARSEGRID_CHUNK_COPY_HPP_ |
10 | |
11 | #if !defined(__NVCC__) || defined(CUDA_ON_CPU) |
12 | #include <Vc/Vc> |
13 | #endif |
14 | #include "util/mathutil.hpp" |
15 | |
16 | |
17 | /*! \brief Check if the point in the chunk exist |
18 | * |
19 | * \param h header |
20 | * \param sub_id index of the sub-domain |
21 | * |
22 | * \return true if exist, false if does not |
23 | * |
24 | */ |
25 | template<typename headerType> |
26 | inline bool exist_sub(headerType & h, int sub_id) |
27 | { |
28 | return h.mask[sub_id]; |
29 | } |
30 | |
31 | template<unsigned int v> |
32 | struct exist_sub_v_impl |
33 | { |
34 | typedef unsigned char type; |
35 | |
36 | template<typename headerType> |
37 | static inline void exist(headerType & h, int sub_id, unsigned char * pmask) |
38 | { |
39 | pmask[0] = h.mask[sub_id]; |
40 | } |
41 | }; |
42 | |
43 | template<> |
44 | struct exist_sub_v_impl<2> |
45 | { |
46 | typedef unsigned short type; |
47 | |
48 | template<typename headerType> |
49 | static inline void exist(headerType & h, int sub_id, unsigned char * pmask) |
50 | { |
51 | pmask[0] = h.mask[sub_id]; |
52 | pmask[1] = h.mask[sub_id+1]; |
53 | } |
54 | }; |
55 | |
56 | template<> |
57 | struct exist_sub_v_impl<4> |
58 | { |
59 | typedef unsigned int type; |
60 | |
61 | template<typename headerType> |
62 | static inline void exist(headerType & h, int sub_id, unsigned char * pmask) |
63 | { |
64 | pmask[0] = h.mask[sub_id]; |
65 | pmask[1] = h.mask[sub_id+1]; |
66 | pmask[2] = h.mask[sub_id+2]; |
67 | pmask[3] = h.mask[sub_id+3]; |
68 | } |
69 | }; |
70 | |
71 | template<> |
72 | struct exist_sub_v_impl<8> |
73 | { |
74 | typedef unsigned long int type; |
75 | |
76 | template<typename headerType> |
77 | static inline void exist(headerType & h, int sub_id, unsigned char * pmask) |
78 | { |
79 | pmask[0] = h.mask[sub_id]; |
80 | pmask[1] = h.mask[sub_id+1]; |
81 | pmask[2] = h.mask[sub_id+2]; |
82 | pmask[3] = h.mask[sub_id+3]; |
83 | pmask[4] = h.mask[sub_id+4]; |
84 | pmask[5] = h.mask[sub_id+5]; |
85 | pmask[6] = h.mask[sub_id+6]; |
86 | pmask[7] = h.mask[sub_id+7]; |
87 | } |
88 | }; |
89 | |
90 | /*! \brief Check if the point in the chunk exist (Vectorial form) |
91 | * |
92 | * \param h header |
93 | * \param sub_id index of the sub-domain |
94 | * |
95 | * \return true if exist, false if does not |
96 | * |
97 | */ |
98 | template<unsigned int v, typename headerType> |
99 | inline void exist_sub_v(headerType & h, int sub_id, unsigned char * pmask) |
100 | { |
101 | exist_sub_v_impl<v>::exist(h,sub_id,pmask); |
102 | } |
103 | |
104 | |
105 | //! Copy block in 3D |
106 | template<int layout_type, int prop, int stencil_size ,typename chunking,bool is_cross> |
107 | struct copy_xyz |
108 | { |
109 | template<unsigned int N1, typename T, typename headerType, typename chunkType> |
110 | inline static void copy(T ptr[N1], unsigned char mask[N1], headerType & h , const chunkType & chunk) |
111 | { |
112 | int s = stencil_size + stencil_size*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*stencil_size) + |
113 | stencil_size*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*stencil_size)*(boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value+2*stencil_size); |
114 | |
115 | int s2 = 0; |
116 | |
117 | for (int v = 0 ; v < boost::mpl::at<typename chunking::type,boost::mpl::int_<2>>::type::value ; v++) |
118 | { |
119 | for (int j = 0 ; j < boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value ; j++) |
120 | { |
121 | for (int k = 0 ; k < boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value ; k++) |
122 | { |
123 | ptr[s] = chunk[s2]; |
124 | mask[s] = exist_sub(h,s2); |
125 | |
126 | s++; |
127 | s2++; |
128 | } |
129 | s += 2*stencil_size; |
130 | } |
131 | s+= 2*stencil_size*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*stencil_size); |
132 | } |
133 | } |
134 | |
135 | template<unsigned int N1, typename T, typename chunkType> |
136 | inline static void store(T ptr[N1] , chunkType & chunk) |
137 | { |
138 | int s2 = 0; |
139 | |
140 | for (int v = 0 ; v < boost::mpl::at<typename chunking::type,boost::mpl::int_<2>>::type::value ; v++) |
141 | { |
142 | for (int j = 0 ; j < boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value ; j++) |
143 | { |
144 | for (int k = 0 ; k < boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value ; k++) |
145 | { |
146 | chunk[s2] = ptr[s2]; |
147 | |
148 | s2++; |
149 | } |
150 | } |
151 | } |
152 | } |
153 | }; |
154 | |
155 | template<unsigned int i> |
156 | struct multi_mask |
157 | {}; |
158 | |
159 | template<> |
160 | struct multi_mask<1> |
161 | { |
162 | typedef unsigned char type; |
163 | }; |
164 | |
165 | template<> |
166 | struct multi_mask<2> |
167 | { |
168 | typedef unsigned short int type; |
169 | }; |
170 | |
171 | template<> |
172 | struct multi_mask<4> |
173 | { |
174 | typedef unsigned int type; |
175 | }; |
176 | |
177 | template<> |
178 | struct multi_mask<8> |
179 | { |
180 | typedef unsigned long int type; |
181 | }; |
182 | |
183 | template<> |
184 | struct multi_mask<16> |
185 | { |
186 | typedef unsigned long int type; |
187 | }; |
188 | |
189 | #ifndef __NVCC__ |
190 | |
191 | //! Copy block in 3D vectorized |
192 | template<int prop, int stencil_size ,typename chunking,bool is_cross> |
193 | struct copy_xyz<1,prop,stencil_size,chunking,is_cross> |
194 | { |
195 | template<unsigned int N1, typename T, typename headerType, typename chunkType> |
196 | inline static void copy(T ptr[N1], unsigned char mask[N1], headerType & h , const chunkType & chunk) |
197 | { |
198 | int s = stencil_size + stencil_size*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*stencil_size) + |
199 | stencil_size*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*stencil_size)*(boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value+2*stencil_size); |
200 | |
201 | typedef boost::mpl::int_<boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value * sizeof(chunk[0]) / |
202 | Vc::float_v::Size / sizeof(float)> n_it_lead; |
203 | |
204 | int s2 = 0; |
205 | |
206 | for (int v = 0 ; v < boost::mpl::at<typename chunking::type,boost::mpl::int_<2>>::type::value ; v++) |
207 | { |
208 | for (int j = 0 ; j < boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value ; j++) |
209 | { |
210 | for (int k = 0 ; k < n_it_lead::value ; k+=4) |
211 | { |
212 | exist_sub_v<Vc::float_v::Size / (sizeof(chunk[0])/sizeof(float))>(h,s2,&mask[s]); |
213 | exist_sub_v<Vc::float_v::Size / (sizeof(chunk[0])/sizeof(float))>(h,s2+Vc::float_v::Size / (sizeof(chunk[0])/sizeof(float)),&mask[s+Vc::float_v::Size / (sizeof(chunk[0])/sizeof(float))]); |
214 | exist_sub_v<Vc::float_v::Size / (sizeof(chunk[0])/sizeof(float))>(h,s2+2*Vc::float_v::Size / (sizeof(chunk[0])/sizeof(float)),&mask[s+2*Vc::float_v::Size / (sizeof(chunk[0])/sizeof(float))]); |
215 | exist_sub_v<Vc::float_v::Size / (sizeof(chunk[0])/sizeof(float))>(h,s2+3*Vc::float_v::Size / (sizeof(chunk[0])/sizeof(float)),&mask[s+3*Vc::float_v::Size / (sizeof(chunk[0])/sizeof(float))]); |
216 | |
217 | Vc::float_v tmp = Vc::float_v((float *)&chunk[s2],Vc::Aligned); |
218 | Vc::float_v tmp2 = Vc::float_v((float *)&chunk[s2+Vc::float_v::Size / (sizeof(chunk[0])/sizeof(float))],Vc::Aligned); |
219 | Vc::float_v tmp3 = Vc::float_v((float *)&chunk[s2+2*Vc::float_v::Size / (sizeof(chunk[0])/sizeof(float))],Vc::Aligned); |
220 | Vc::float_v tmp4 = Vc::float_v((float *)&chunk[s2+3*Vc::float_v::Size / (sizeof(chunk[0])/sizeof(float))],Vc::Aligned); |
221 | tmp.store((float *)&ptr[s],Vc::Unaligned); |
222 | tmp2.store((float *)&ptr[s+Vc::float_v::Size / (sizeof(chunk[0])/sizeof(float))],Vc::Unaligned); |
223 | tmp3.store((float *)&ptr[s+2*Vc::float_v::Size / (sizeof(chunk[0])/sizeof(float))],Vc::Unaligned); |
224 | tmp4.store((float *)&ptr[s+3*Vc::float_v::Size / (sizeof(chunk[0])/sizeof(float))],Vc::Unaligned); |
225 | |
226 | s += 4*Vc::float_v::Size / (sizeof(chunk[0])/sizeof(float)); |
227 | s2 += 4*Vc::float_v::Size / (sizeof(chunk[0])/sizeof(float)); |
228 | } |
229 | s += 2*stencil_size ; |
230 | } |
231 | s+= 2*stencil_size*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*stencil_size); |
232 | } |
233 | } |
234 | |
235 | template<unsigned int N1, typename T, typename chunkType> |
236 | inline static void store(T ptr[N1] , chunkType & chunk) |
237 | { |
238 | typedef boost::mpl::int_<boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value * sizeof(chunk[0]) / |
239 | Vc::float_v::Size / sizeof(float)> n_it_lead; |
240 | |
241 | int s2 = 0; |
242 | |
243 | for (int v = 0 ; v < boost::mpl::at<typename chunking::type,boost::mpl::int_<2>>::type::value ; v++) |
244 | { |
245 | for (int j = 0 ; j < boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value ; j++) |
246 | { |
247 | for (int k = 0 ; k < n_it_lead::value ; k += 4) |
248 | { |
249 | Vc::float_v tmp = Vc::float_v((float *)&ptr[s2],Vc::Aligned); |
250 | Vc::float_v tmp1 = Vc::float_v((float *)&ptr[s2+1*Vc::float_v::Size / (sizeof(chunk[0])/sizeof(float))],Vc::Aligned); |
251 | Vc::float_v tmp2 = Vc::float_v((float *)&ptr[s2+2*Vc::float_v::Size / (sizeof(chunk[0])/sizeof(float))],Vc::Aligned); |
252 | Vc::float_v tmp3 = Vc::float_v((float *)&ptr[s2+3*Vc::float_v::Size / (sizeof(chunk[0])/sizeof(float))],Vc::Aligned); |
253 | |
254 | tmp.store((float *)&chunk[s2],Vc::Aligned); |
255 | tmp1.store((float *)&chunk[s2+1*Vc::float_v::Size / (sizeof(chunk[0])/sizeof(float))],Vc::Aligned); |
256 | tmp2.store((float *)&chunk[s2+2*Vc::float_v::Size / (sizeof(chunk[0])/sizeof(float))],Vc::Aligned); |
257 | tmp3.store((float *)&chunk[s2+3*Vc::float_v::Size / (sizeof(chunk[0])/sizeof(float))],Vc::Aligned); |
258 | |
259 | s2 += 4*Vc::float_v::Size / (sizeof(chunk[0])/sizeof(float)); |
260 | } |
261 | } |
262 | } |
263 | } |
264 | }; |
265 | |
266 | #endif |
267 | |
268 | //! Copy XY surface in 3D |
269 | template<int layout_type, int prop,int stencil_size, typename chunking,bool is_cross> |
270 | struct copy_xy_3 |
271 | { |
272 | template<unsigned int i_src, unsigned int i_dest, unsigned int N1, typename T, typename headerType, typename chunkType> |
273 | inline static void copy(T ptr[N1], unsigned char mask[N1], headerType & h , const chunkType & chunk) |
274 | { |
275 | int s = stencil_size + stencil_size*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*stencil_size) + |
276 | i_dest*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*stencil_size)*(boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value+2*stencil_size); |
277 | |
278 | for (int v = 0 ; v < stencil_size ; v++) |
279 | { |
280 | for (int j = 0 ; j < boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value ; j++) |
281 | { |
282 | for (int k = 0 ; k < boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value ; k++) |
283 | { |
284 | const int id = Lin_vmpl<typename chunking::type>(k,j,i_src+v); |
285 | |
286 | ptr[s] = chunk.template get<prop>()[id]; |
287 | mask[s] = exist_sub(h,id); |
288 | |
289 | s++; |
290 | } |
291 | s += 2*stencil_size; |
292 | } |
293 | |
294 | s+= 2*stencil_size*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*stencil_size); |
295 | } |
296 | } |
297 | |
298 | template<unsigned int i_dest, unsigned int N1> |
299 | inline static void mask_null(unsigned char mask[N1]) |
300 | { |
301 | int s = stencil_size + stencil_size*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*stencil_size) + |
302 | i_dest*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*stencil_size)*(boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value+2*stencil_size); |
303 | |
304 | for (int v = 0 ; v < stencil_size ; v++) |
305 | { |
306 | for (int j = 0 ; j < boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value ; j++) |
307 | { |
308 | for (int k = 0 ; k < boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value ; k++) |
309 | { |
310 | mask[s] = 0; |
311 | |
312 | s++; |
313 | } |
314 | s += 2*stencil_size; |
315 | } |
316 | |
317 | s+= 2*stencil_size*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*stencil_size); |
318 | } |
319 | } |
320 | }; |
321 | |
322 | #ifndef __NVCC__ |
323 | |
324 | //! Copy XY surface in 3D |
325 | template<int prop,int stencil_size, typename chunking,bool is_cross> |
326 | struct copy_xy_3<1,prop,stencil_size,chunking,is_cross> |
327 | { |
328 | template<unsigned int i_src, unsigned int i_dest, unsigned int N1, typename T, typename headerType, typename chunkType> |
329 | inline static void copy(T ptr[N1], unsigned char mask[N1], headerType & h , const chunkType & chunk) |
330 | { |
331 | int s = stencil_size + stencil_size*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*stencil_size) + |
332 | i_dest*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*stencil_size)*(boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value+2*stencil_size); |
333 | |
334 | int s2 = i_src*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value)*(boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value); |
335 | |
336 | typedef boost::mpl::int_<boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value * sizeof(chunk.template get<prop>()[0]) / |
337 | Vc::float_v::Size / sizeof(float)> n_it_lead; |
338 | |
339 | for (int v = 0 ; v < stencil_size ; v++) |
340 | { |
341 | for (int j = 0 ; j < boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value ; j++) |
342 | { |
343 | for (int k = 0 ; k < n_it_lead::value ; k+=4) |
344 | { |
345 | exist_sub_v<Vc::float_v::Size / (sizeof(chunk.template get<prop>()[0])/sizeof(float))>(h,s2,&mask[s]); |
346 | exist_sub_v<Vc::float_v::Size / (sizeof(chunk.template get<prop>()[0])/sizeof(float))>(h,s2+1*Vc::float_v::Size / (sizeof(chunk.template get<prop>()[0])/sizeof(float)),&mask[s+1*Vc::float_v::Size / (sizeof(chunk.template get<prop>()[0])/sizeof(float))]); |
347 | exist_sub_v<Vc::float_v::Size / (sizeof(chunk.template get<prop>()[0])/sizeof(float))>(h,s2+2*Vc::float_v::Size / (sizeof(chunk.template get<prop>()[0])/sizeof(float)),&mask[s+2*Vc::float_v::Size / (sizeof(chunk.template get<prop>()[0])/sizeof(float))]); |
348 | exist_sub_v<Vc::float_v::Size / (sizeof(chunk.template get<prop>()[0])/sizeof(float))>(h,s2+3*Vc::float_v::Size / (sizeof(chunk.template get<prop>()[0])/sizeof(float)),&mask[s+3*Vc::float_v::Size / (sizeof(chunk.template get<prop>()[0])/sizeof(float))]); |
349 | |
350 | |
351 | Vc::float_v tmp = Vc::float_v((float *)&chunk.template get<prop>()[s2],Vc::Unaligned); |
352 | Vc::float_v tmp1 = Vc::float_v((float *)&chunk.template get<prop>()[s2+1*Vc::float_v::Size / (sizeof(chunk.template get<prop>()[0])/sizeof(float))],Vc::Unaligned); |
353 | Vc::float_v tmp2 = Vc::float_v((float *)&chunk.template get<prop>()[s2+2*Vc::float_v::Size / (sizeof(chunk.template get<prop>()[0])/sizeof(float))],Vc::Unaligned); |
354 | Vc::float_v tmp3 = Vc::float_v((float *)&chunk.template get<prop>()[s2+3*Vc::float_v::Size / (sizeof(chunk.template get<prop>()[0])/sizeof(float))],Vc::Unaligned); |
355 | tmp.store((float *)&ptr[s],Vc::Unaligned); |
356 | tmp1.store((float *)&ptr[s+1*Vc::float_v::Size / (sizeof(chunk.template get<prop>()[0])/sizeof(float))],Vc::Unaligned); |
357 | tmp2.store((float *)&ptr[s+2*Vc::float_v::Size / (sizeof(chunk.template get<prop>()[0])/sizeof(float))],Vc::Unaligned); |
358 | tmp3.store((float *)&ptr[s+3*Vc::float_v::Size / (sizeof(chunk.template get<prop>()[0])/sizeof(float))],Vc::Unaligned); |
359 | |
360 | s += 4*Vc::float_v::Size / (sizeof(chunk.template get<prop>()[0])/sizeof(float)); |
361 | s2 += 4*Vc::float_v::Size / (sizeof(chunk.template get<prop>()[0])/sizeof(float)); |
362 | } |
363 | s += 2*stencil_size; |
364 | |
365 | } |
366 | |
367 | s+= 2*stencil_size*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*stencil_size); |
368 | } |
369 | } |
370 | |
371 | template<unsigned int i_dest, unsigned int N1> |
372 | inline static void mask_null(unsigned char mask[N1]) |
373 | { |
374 | int s = stencil_size + stencil_size*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*stencil_size) + |
375 | i_dest*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*stencil_size)*(boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value+2*stencil_size); |
376 | |
377 | for (int v = 0 ; v < stencil_size ; v++) |
378 | { |
379 | for (int j = 0 ; j < boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value ; j++) |
380 | { |
381 | for (int k = 0 ; k < boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value ; k++) |
382 | { |
383 | mask[s] = 0; |
384 | |
385 | s++; |
386 | } |
387 | s += 2*stencil_size; |
388 | } |
389 | |
390 | s+= 2*stencil_size*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*stencil_size); |
391 | } |
392 | } |
393 | }; |
394 | |
395 | #endif |
396 | |
397 | //! Copy XZ surface in 3D |
398 | template<int layout_type, int prop, int stencil_size, typename chunking,bool is_cross> |
399 | struct copy_xz_3 |
400 | { |
401 | template<unsigned int j_src, unsigned int j_dest, unsigned int N1, typename T, typename headerType , typename chunkType> |
402 | inline static void copy(T ptr[N1], unsigned char mask[N1], headerType & h , const chunkType & chunk) |
403 | { |
404 | int s = stencil_size + j_dest*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*stencil_size) + |
405 | stencil_size*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*stencil_size)*(boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value+2*stencil_size); |
406 | |
407 | for (int v = 0 ; v < stencil_size ; v++) |
408 | { |
409 | for (int i = 0 ; i < boost::mpl::at<typename chunking::type,boost::mpl::int_<2>>::type::value ; i++) |
410 | { |
411 | for (int k = 0 ; k < boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value ; k++) |
412 | { |
413 | const int id = Lin_vmpl<typename chunking::type>(k,j_src+v,i); |
414 | |
415 | ptr[s] = chunk.template get<prop>()[id]; |
416 | mask[s] = exist_sub(h,id); |
417 | |
418 | s++; |
419 | } |
420 | |
421 | s += (boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*stencil_size)*(boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value + 2*stencil_size) - boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value; |
422 | } |
423 | |
424 | s += (boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value + 2*stencil_size) - boost::mpl::at<typename chunking::type,boost::mpl::int_<2>>::type::value*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*stencil_size)*(boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value+2*stencil_size); |
425 | } |
426 | } |
427 | }; |
428 | |
429 | //! Copy XZ surface in 3D |
430 | template<int layout_type, int prop, typename chunking,bool is_cross> |
431 | struct copy_xz_3<layout_type,prop,1,chunking,is_cross> |
432 | { |
433 | template<unsigned int j_src, unsigned int j_dest, unsigned int N1, typename T, typename headerType , typename chunkType> |
434 | inline static void copy(T ptr[N1], unsigned char mask[N1], headerType & h , const chunkType & chunk) |
435 | { |
436 | int s = 1 + j_dest*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*1) + |
437 | 1*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*1)*(boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value+2*1); |
438 | |
439 | for (int i = 0 ; i < boost::mpl::at<typename chunking::type,boost::mpl::int_<2>>::type::value ; i++) |
440 | { |
441 | for (int k = 0 ; k < boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value ; k++) |
442 | { |
443 | const int id = Lin_vmpl<typename chunking::type>(k,j_src,i); |
444 | |
445 | ptr[s] = chunk.template get<prop>()[id]; |
446 | mask[s] = exist_sub(h,id); |
447 | |
448 | s++; |
449 | } |
450 | |
451 | s += 1 * (boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*1)*(boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value + 2*1) - boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value; |
452 | } |
453 | } |
454 | |
455 | template<unsigned int j_dest, unsigned int N1> |
456 | inline static void mask_null(unsigned char mask[N1]) |
457 | { |
458 | int s = 1 + j_dest*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*1) + |
459 | 1*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*1)*(boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value+2*1); |
460 | |
461 | for (int i = 0 ; i < boost::mpl::at<typename chunking::type,boost::mpl::int_<2>>::type::value ; i++) |
462 | { |
463 | for (int k = 0 ; k < boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value ; k++) |
464 | { |
465 | mask[s] = 0; |
466 | |
467 | s++; |
468 | } |
469 | |
470 | s += 1 * (boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*1)*(boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value + 2*1) - boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value; |
471 | } |
472 | } |
473 | }; |
474 | |
475 | //! Copy YZ surface in 3D |
476 | template<int layout_type, int prop, int stencil_size, typename chunking,bool is_cross> |
477 | struct copy_yz_3 |
478 | { |
479 | template<unsigned int k_src, unsigned int k_dest, unsigned int N1, typename T, typename headerType , typename chunkType> |
480 | inline static void copy(T ptr[N1], unsigned char mask[N1], headerType & h , const chunkType & chunk) |
481 | { |
482 | int s = k_dest + stencil_size*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*stencil_size) + |
483 | stencil_size*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*stencil_size)*(boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value+2*stencil_size); |
484 | |
485 | for (int v = 0 ; v < stencil_size ; v++) |
486 | { |
487 | for (int i = 0 ; i < boost::mpl::at<typename chunking::type,boost::mpl::int_<2>>::type::value ; i++) |
488 | { |
489 | for (int j = 0 ; j < boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value ; j++) |
490 | { |
491 | const int id = Lin_vmpl<typename chunking::type>(k_src+v,j,i); |
492 | |
493 | ptr[s] = chunk.template get<prop>()[id]; |
494 | mask[s] = exist_sub(h,id); |
495 | |
496 | s += (boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*stencil_size); |
497 | } |
498 | |
499 | s += (boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*stencil_size)*(boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value + 2*stencil_size) - boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*stencil_size); |
500 | } |
501 | |
502 | s += 1 - boost::mpl::at<typename chunking::type,boost::mpl::int_<2>>::type::value*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*stencil_size)*(boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value+2*stencil_size); |
503 | |
504 | } |
505 | } |
506 | }; |
507 | |
508 | //! Copy YZ surface in 3D |
509 | template<int layout_type, int prop, typename chunking,bool is_cross> |
510 | struct copy_yz_3<layout_type,prop,1,chunking,is_cross> |
511 | { |
512 | template<unsigned int k_src, unsigned int k_dest, unsigned int N1, typename T, typename headerType, typename chunkType> |
513 | inline static void copy(T ptr[N1], unsigned char mask[N1], headerType & h , const chunkType & chunk) |
514 | { |
515 | int s = k_dest + 1*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*1) + |
516 | 1*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*1)*(boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value+2*1); |
517 | |
518 | for (int i = 0 ; i < boost::mpl::at<typename chunking::type,boost::mpl::int_<2>>::type::value ; i++) |
519 | { |
520 | for (int j = 0 ; j < boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value ; j++) |
521 | { |
522 | const int id = Lin_vmpl<typename chunking::type>(k_src,j,i); |
523 | |
524 | ptr[s] = chunk.template get<prop>()[id]; |
525 | mask[s] = exist_sub(h,id); |
526 | |
527 | s += (boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*1); |
528 | } |
529 | |
530 | s += 1 * (boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*1)*(boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value + 2*1) - boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*1); |
531 | } |
532 | } |
533 | |
534 | template<unsigned int k_dest, unsigned int N1> |
535 | inline static void mask_null(unsigned char mask[N1]) |
536 | { |
537 | int s = k_dest + 1*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*1) + |
538 | 1*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*1)*(boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value+2*1); |
539 | |
540 | for (int i = 0 ; i < boost::mpl::at<typename chunking::type,boost::mpl::int_<2>>::type::value ; i++) |
541 | { |
542 | for (int j = 0 ; j < boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value ; j++) |
543 | { |
544 | mask[s] = 0; |
545 | |
546 | s += (boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*1); |
547 | } |
548 | |
549 | s += 1 * (boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*1)*(boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value + 2*1) - boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*1); |
550 | } |
551 | } |
552 | }; |
553 | |
554 | //! Copy x edge in 3D |
555 | template<int layout_type, int prop, int stencil_size, typename chunking,bool is_cross> |
556 | struct copy_x_3 |
557 | { |
558 | template<unsigned int i_src, unsigned int i_dest,unsigned int j_src, unsigned int j_dest , unsigned int N1, typename T, typename chunkType> |
559 | inline static void copy(T ptr[N1], const chunkType & chunk) |
560 | { |
561 | int s = stencil_size + j_dest*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*stencil_size) + |
562 | i_dest*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*stencil_size)*(boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value+2*stencil_size); |
563 | |
564 | for (int v1 = 0 ; v1 < stencil_size ; v1++) |
565 | { |
566 | for (int v2 = 0 ; v2 < stencil_size ; v2++) |
567 | { |
568 | for (int k = 0 ; k < boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value ; k++) |
569 | { |
570 | ptr[s] = chunk.template get<prop>()[Lin_vmpl<typename chunking::type>(k,j_src+v2,i_src+v1)]; |
571 | |
572 | s++; |
573 | } |
574 | |
575 | s+= 2*stencil_size; |
576 | } |
577 | |
578 | s+= (boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*stencil_size)*(boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value+2*stencil_size) - stencil_size*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*stencil_size); |
579 | } |
580 | } |
581 | }; |
582 | |
583 | //! Copy x edge in 3D |
584 | template<int layout_type, int prop, typename chunking,bool is_cross> |
585 | struct copy_x_3<layout_type,prop,1,chunking,is_cross> |
586 | { |
587 | template<unsigned int i_src, unsigned int i_dest,unsigned int j_src, unsigned int j_dest , unsigned int N1, typename T, typename chunkType> |
588 | inline static void copy(T ptr[N1], const chunkType & chunk) |
589 | { |
590 | int s = 1 + j_dest*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*1) + |
591 | i_dest*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*1)*(boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value+2*1); |
592 | |
593 | |
594 | for (int k = 0 ; k < boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value ; k++) |
595 | { |
596 | ptr[s] = chunk.template get<prop>()[Lin_vmpl<typename chunking::type>(k,j_src,i_src)]; |
597 | |
598 | s++; |
599 | } |
600 | } |
601 | }; |
602 | |
603 | //! Copy x edge in 3D |
604 | template<int layout_type, int prop, int stencil_size, typename chunking> |
605 | struct copy_x_3<layout_type,prop,stencil_size,chunking,true> |
606 | { |
607 | template<unsigned int i_src, unsigned int i_dest,unsigned int j_src, unsigned int j_dest , unsigned int N1, typename T, typename chunkType> |
608 | inline static void copy(T ptr[N1], const chunkType & chunk) |
609 | {} |
610 | }; |
611 | |
612 | //! Copy y edge in 3D |
613 | template<int layout_type, int prop, int stencil_size, typename chunking,bool is_cross> |
614 | struct copy_y_3 |
615 | { |
616 | template<unsigned int i_src, unsigned int i_dest,unsigned int k_src, unsigned int k_dest , unsigned int N1, typename T, typename chunkType> |
617 | inline static void copy(T ptr[N1], const chunkType & chunk) |
618 | { |
619 | int s = k_dest + stencil_size*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*stencil_size) + |
620 | i_dest*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*stencil_size)*(boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value+2*stencil_size); |
621 | |
622 | for (int v1 = 0 ; v1 < stencil_size ; v1++) |
623 | { |
624 | for (int j = 0 ; j < boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value ; j++) |
625 | { |
626 | for (int v2 = 0 ; v2 < stencil_size ; v2++) |
627 | { |
628 | ptr[s] = chunk.template get<prop>()[Lin_vmpl<typename chunking::type>(k_src+v2,j,i_src+v1)]; |
629 | s+= 1; |
630 | } |
631 | |
632 | s += (boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*stencil_size) - stencil_size; |
633 | } |
634 | |
635 | s += (boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*stencil_size)*(boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value+2*stencil_size) - boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*stencil_size); |
636 | } |
637 | } |
638 | }; |
639 | |
640 | //! Copy y edge in 3D |
641 | template<int layout_type, int prop, typename chunking,bool is_cross> |
642 | struct copy_y_3<layout_type,prop,1,chunking,is_cross> |
643 | { |
644 | template<unsigned int i_src, unsigned int i_dest,unsigned int k_src, unsigned int k_dest , unsigned int N1, typename T, typename chunkType> |
645 | inline static void copy(T ptr[N1], const chunkType & chunk) |
646 | { |
647 | int s = k_dest + 1*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*1) + |
648 | i_dest*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*1)*(boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value+2*1); |
649 | |
650 | |
651 | for (int j = 0 ; j < boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value ; j++) |
652 | { |
653 | ptr[s] = chunk.template get<prop>()[Lin_vmpl<typename chunking::type>(k_src,j,i_src)]; |
654 | |
655 | s += (boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*1); |
656 | } |
657 | } |
658 | }; |
659 | |
660 | //! Copy y edge in 3D |
661 | template<int layout_type, int prop, int stencil_size, typename chunking> |
662 | struct copy_y_3<layout_type,prop,stencil_size,chunking,true> |
663 | { |
664 | template<unsigned int i_src, unsigned int i_dest,unsigned int k_src, unsigned int k_dest , unsigned int N1, typename T, typename chunkType> |
665 | inline static void copy(T ptr[N1], const chunkType & chunk) |
666 | {} |
667 | }; |
668 | |
669 | |
670 | //! Copy z edge in 3D |
671 | template<int layout_type, int prop, int stencil_size, typename chunking,bool is_cross> |
672 | struct copy_z_3 |
673 | { |
674 | template<unsigned int j_src, unsigned int j_dest,unsigned int k_src, unsigned int k_dest , unsigned int N1, typename T, typename chunkType> |
675 | inline static void copy(T ptr[N1], const chunkType & chunk) |
676 | { |
677 | int s = k_dest + j_dest*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*stencil_size) + |
678 | stencil_size*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*stencil_size)*(boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value+2*stencil_size); |
679 | |
680 | |
681 | for (int i = 0 ; i < boost::mpl::at<typename chunking::type,boost::mpl::int_<2>>::type::value ; i++) |
682 | { |
683 | for (int v1 = 0 ; v1 < stencil_size ; v1++) |
684 | { |
685 | for (int v2 = 0 ; v2 < stencil_size ; v2++) |
686 | { |
687 | ptr[s] = chunk.template get<prop>()[Lin_vmpl<typename chunking::type>(k_src+v2,j_src+v1,i)]; |
688 | |
689 | s++; |
690 | } |
691 | |
692 | s+= 2*stencil_size; |
693 | } |
694 | |
695 | s += (boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*stencil_size)*(boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value+2*stencil_size); |
696 | } |
697 | } |
698 | }; |
699 | |
700 | //! Copy z edge in 3D |
701 | template<int layout_type, int prop, typename chunking,bool is_cross> |
702 | struct copy_z_3<layout_type,prop,1,chunking,is_cross> |
703 | { |
704 | template<unsigned int j_src, unsigned int j_dest,unsigned int k_src, unsigned int k_dest , unsigned int N1, typename T, typename chunkType> |
705 | inline static void copy(T ptr[N1], const chunkType & chunk) |
706 | { |
707 | int s = k_dest + j_dest*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*1) + |
708 | 1*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*1)*(boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value+2*1); |
709 | |
710 | |
711 | for (int i = 0 ; i < boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value ; i++) |
712 | { |
713 | ptr[s] = chunk.template get<prop>()[Lin_vmpl<typename chunking::type>(k_src,j_src,i)]; |
714 | |
715 | s += (boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*1)*(boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value+2*1); |
716 | } |
717 | } |
718 | }; |
719 | |
720 | //! Copy z edge in 3D |
721 | template<int layout_type, int prop, int stencil_size, typename chunking> |
722 | struct copy_z_3<layout_type,prop,stencil_size,chunking,true> |
723 | { |
724 | template<unsigned int j_src, unsigned int j_dest,unsigned int k_src, unsigned int k_dest , unsigned int N1, typename T, typename chunkType> |
725 | inline static void copy(T ptr[N1], const chunkType & chunk) |
726 | {} |
727 | }; |
728 | |
729 | //! Copy point in 3D |
730 | template<int layout_type, int prop, int stencil_size, typename chunking,bool is_cross> |
731 | struct copy_corner_3 |
732 | { |
733 | template<unsigned int i_src, unsigned int i_dest,unsigned int j_src, unsigned int j_dest,unsigned int k_src, unsigned int k_dest , unsigned int N1, typename T, typename chunkType> |
734 | inline static void copy(T ptr[N1], const chunkType & chunk) |
735 | { |
736 | int s = k_dest + j_dest*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*stencil_size) + |
737 | i_dest*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*stencil_size)*(boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value+2*stencil_size); |
738 | |
739 | |
740 | for (int i = 0 ; i < stencil_size ; i++) |
741 | { |
742 | for (int j = 0 ; j < stencil_size ; j++) |
743 | { |
744 | for (int k = 0 ; k < stencil_size ; k++) |
745 | { |
746 | ptr[s] = chunk.template get<prop>()[Lin_vmpl<typename chunking::type>(k_src+k,j_src+j,i_src+i)]; |
747 | |
748 | s++; |
749 | } |
750 | |
751 | s += (boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*stencil_size) - stencil_size; |
752 | } |
753 | |
754 | s += (boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*stencil_size)*(boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value+2*stencil_size) - stencil_size*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*stencil_size); |
755 | } |
756 | } |
757 | }; |
758 | |
759 | //! Copy point in 3D |
760 | template<int layout_type, int prop, typename chunking,bool is_cross> |
761 | struct copy_corner_3<layout_type,prop,1,chunking,is_cross> |
762 | { |
763 | template<unsigned int i_src, unsigned int i_dest,unsigned int j_src, unsigned int j_dest,unsigned int k_src, unsigned int k_dest , unsigned int N1, typename T, typename chunkType> |
764 | inline static void copy(T ptr[N1], const chunkType & chunk) |
765 | { |
766 | int s = k_dest + j_dest*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*1) + |
767 | i_dest*(boost::mpl::at<typename chunking::type,boost::mpl::int_<0>>::type::value+2*1)*(boost::mpl::at<typename chunking::type,boost::mpl::int_<1>>::type::value+2*1); |
768 | |
769 | |
770 | ptr[s] = chunk.template get<prop>()[Lin_vmpl<typename chunking::type>(k_src,j_src,i_src)]; |
771 | } |
772 | }; |
773 | |
774 | //! Copy point in 3D |
775 | template<int layout_type, int prop, int stencil_size, typename chunking> |
776 | struct copy_corner_3<layout_type,prop,stencil_size,chunking,true> |
777 | { |
778 | template<unsigned int i_src, unsigned int i_dest,unsigned int j_src, unsigned int j_dest,unsigned int k_src, unsigned int k_dest , unsigned int N1, typename T, typename chunkType> |
779 | inline static void copy(T ptr[N1], const chunkType & chunk) |
780 | {} |
781 | }; |
782 | |
783 | ///////////////////////////////////////////////////////////////////////// Chunk missalignment mapping |
784 | |
785 | template<unsigned int dim> |
786 | struct key_int |
787 | { |
788 | unsigned char i; |
789 | grid_key_dx<dim,long int> k; |
790 | }; |
791 | |
792 | /*! to move chunks from one Sparse grid to another SparseGrid chunk can be miss-aligned this function create a map between |
793 | * the missaligned source and the destination chunks |
794 | * |
795 | * For example in an 8x8 chunk in which the chunk in the source sparse grid are shifted by two from the destination sparse-grid |
796 | * you will get a map like this |
797 | * |
798 | /verbatim |
799 | |
800 | 2 2 3 3 3 3 3 3 |
801 | 2 2 3 3 3 3 3 3 |
802 | 2 2 3 3 3 3 3 3 |
803 | 2 2 3 3 3 3 3 3 |
804 | 0 0 1 1 1 1 1 1 |
805 | 0 0 1 1 1 1 1 1 |
806 | |
807 | /endverbatim |
808 | * |
809 | * and a vk filled with {-1,-1},0 {0,-1},1 {-1,0},2 {0,0},3 |
810 | * |
811 | * in case we are comping a slice some of the index in vk can be filtered out. For example if we are copyng a slice in x = 0 |
812 | * the vector will only have: |
813 | * |
814 | * {-1,-1},0 {-1,0},2 |
815 | * |
816 | */ |
817 | template<unsigned int dim, unsigned int N, typename chunking> |
818 | void construct_chunk_missalign_map(unsigned char miss_al_map[N], |
819 | short int mp_off[N], |
820 | const Box<dim,long int> & b_src, |
821 | const Box<dim,long int> & b_dst, |
822 | openfpm::vector<key_int<dim>> & vk) |
823 | { |
824 | typedef typename vmpl_create_constant<dim,1>::type one_vmpl; |
825 | |
826 | get_block_sizes<dim,0,typename chunking::type,one_vmpl> gbs; |
827 | boost::mpl::for_each_ref< boost::mpl::range_c<int,0,dim> >(gbs); |
828 | |
829 | grid_key_dx<dim> k_src = b_src.getKP1(); |
830 | grid_key_dx<dim> k_dst = b_dst.getKP1(); |
831 | |
832 | grid_key_dx<dim,long int> kl1; |
833 | grid_key_dx<dim,long int> kl2; |
834 | |
835 | // shift the key |
836 | key_shift<dim,chunking>::shift(k_src,kl1); |
837 | |
838 | // shift the key |
839 | key_shift<dim,chunking>::shift(k_dst,kl2); |
840 | |
841 | grid_key_dx<dim,long int> key_sub; |
842 | |
843 | for (int i = 0 ; i < dim ; i++) |
844 | { |
845 | key_sub.set_d(i,kl2.get(i) - kl1.get(i) ); |
846 | } |
847 | |
848 | int c_id[openfpm::math::pow(2,dim)]; |
849 | size_t sz[dim]; |
850 | |
851 | for (int i = 0 ; i < dim ; i++) |
852 | {sz[i] = 3;} |
853 | |
854 | grid_sm<dim,void> gcnk(gbs.sz_block); |
855 | grid_key_dx_iterator<dim> it(gcnk); |
856 | while (it.isNext()) |
857 | { |
858 | auto k = it.get(); |
859 | |
860 | int off = 0; |
861 | int bid = 0; |
862 | int stride = 1; |
863 | int stride_off = 1; |
864 | for (int i = 0 ; i < dim ; i++) |
865 | { |
866 | if ((long int)k.get(i) + key_sub.get(i) >= (long int)gbs.sz_block[i]) |
867 | { |
868 | bid += 2*stride; |
869 | |
870 | } |
871 | else if ((long int)k.get(i) + key_sub.get(i) < 0) |
872 | { |
873 | bid += 0*stride; |
874 | |
875 | } |
876 | else |
877 | { |
878 | bid += 1*stride; |
879 | } |
880 | off += (openfpm::math::positive_modulo(key_sub.get(i) + k.get(i),gbs.sz_block[i]))*stride_off; |
881 | stride *= 3; |
882 | stride_off *= gbs.sz_block[i]; |
883 | } |
884 | |
885 | miss_al_map[gcnk.LinId(k)] = bid; |
886 | mp_off[gcnk.LinId(k)] = off; |
887 | |
888 | ++it; |
889 | } |
890 | |
891 | // Filtering: sometime in particular for ghost copy we have a slice to copy, in this case some of the index in the map |
892 | // are never touched so we can eliminate elements in the vk |
893 | |
894 | Box<dim,long int> slice; |
895 | |
896 | // calculate slice |
897 | for (int i = 0 ; i < dim ; i++) |
898 | { |
899 | if (kl1.get(i) + (b_src.getHigh(i) - b_src.getLow(i) + 1) > gbs.sz_block[i]) |
900 | { |
901 | slice.setLow(i,0); |
902 | slice.setHigh(i,gbs.sz_block[i]-1); |
903 | } |
904 | else |
905 | { |
906 | slice.setLow(i,kl1.get(i)); |
907 | slice.setHigh(i,kl1.get(i) + (b_src.getHigh(i) - b_src.getLow(i) + 1)); |
908 | } |
909 | } |
910 | |
911 | key_int<dim> l; |
912 | l.i = 0; |
913 | l.k.zero(); |
914 | vk.add(l); |
915 | |
916 | for (int i = 0 ; i < dim ; i++) |
917 | { |
918 | if (key_sub.get(i) >= 0) |
919 | { |
920 | size_t bord = gbs.sz_block[i] - key_sub.get(i); |
921 | |
922 | if (slice.getLow(i) < bord) |
923 | { |
924 | // set the component |
925 | for (int j = 0 ; j < vk.size() ; j++) |
926 | { |
927 | vk.get(j).k.set_d(i,0); |
928 | } |
929 | if (bord <= slice.getHigh(i) ) |
930 | { |
931 | int n_dup = vk.size(); |
932 | |
933 | // add the other parts with zero |
934 | for (int j = 0 ; j < n_dup ; j++) |
935 | { |
936 | key_int<dim> tmp; |
937 | tmp.k = vk.get(j).k; |
938 | |
939 | tmp.k.set_d(i,1); |
940 | |
941 | vk.add(tmp); |
942 | } |
943 | } |
944 | } |
945 | else |
946 | { |
947 | // set the component |
948 | for (int j = 0 ; j < vk.size() ; j++) |
949 | { |
950 | vk.get(j).k.set_d(i,0); |
951 | } |
952 | } |
953 | } |
954 | else |
955 | { |
956 | size_t bord = -key_sub.get(i); |
957 | |
958 | if (slice.getLow(i) < bord) |
959 | { |
960 | // set the component |
961 | for (int j = 0 ; j < vk.size() ; j++) |
962 | { |
963 | vk.get(j).k.set_d(i,-1); |
964 | } |
965 | if (bord <= slice.getHigh(i) ) |
966 | { |
967 | int n_dup = vk.size(); |
968 | |
969 | // add the other parts with zero |
970 | for (int j = 0 ; j < n_dup ; j++) |
971 | { |
972 | key_int<dim> tmp; |
973 | tmp.k = vk.get(j).k; |
974 | |
975 | tmp.k.set_d(i,0); |
976 | |
977 | vk.add(tmp); |
978 | } |
979 | } |
980 | } |
981 | else |
982 | { |
983 | // set the component |
984 | for (int j = 0 ; j < vk.size() ; j++) |
985 | { |
986 | vk.get(j).k.set_d(i,0); |
987 | } |
988 | } |
989 | } |
990 | |
991 | /* size_t bord = (gbs.sz_block[i] - key_sub.get(i)); |
992 | |
993 | if (slice.getLow(i) < (gbs.sz_block[i] - key_sub.get(i))) |
994 | { |
995 | // set the component |
996 | for (int j = 0 ; j < vk.size() ; j++) |
997 | { |
998 | vk.get(j).k.set_d(i,0); |
999 | } |
1000 | if ((gbs.sz_block[i] - key_sub.get(i)) <= slice.getHigh(i) ) |
1001 | { |
1002 | int n_dup = vk.size(); |
1003 | |
1004 | // add the other parts with zero |
1005 | for (int j = 0 ; j < n_dup ; j++) |
1006 | { |
1007 | key_int<dim> tmp; |
1008 | tmp.k = vk.get(j).k; |
1009 | |
1010 | tmp.k.set_d(i,0); |
1011 | |
1012 | vk.add(tmp); |
1013 | } |
1014 | } |
1015 | } |
1016 | else |
1017 | { |
1018 | // set the component |
1019 | for (int j = 0 ; j < vk.size() ; j++) |
1020 | { |
1021 | vk.get(j).k.set_d(i,0); |
1022 | } |
1023 | }*/ |
1024 | } |
1025 | } |
1026 | |
1027 | |
1028 | template<typename SparseGridType> |
1029 | void copy_remove_to_impl(const SparseGridType & grid_src, |
1030 | SparseGridType & grid_dst, |
1031 | const Box<SparseGridType::dims,long int> & b_src, |
1032 | const Box<SparseGridType::dims,long int> & b_dst) |
1033 | { |
1034 | typedef typename vmpl_create_constant<SparseGridType::dims,1>::type one_vmpl; |
1035 | |
1036 | get_block_sizes<SparseGridType::dims,0,typename SparseGridType::chunking_type::type,one_vmpl> gbs; |
1037 | boost::mpl::for_each_ref< boost::mpl::range_c<int,0,SparseGridType::dims> >(gbs); |
1038 | |
1039 | typedef typename vmpl_reduce_prod<typename SparseGridType::chunking_type::type>::type sizeBlock; |
1040 | |
1041 | unsigned char miss_al_map[sizeBlock::value]; |
1042 | short int mp_off[sizeBlock::value]; |
1043 | |
1044 | openfpm::vector<key_int<SparseGridType::dims>> vk; |
1045 | |
1046 | construct_chunk_missalign_map<SparseGridType::dims, |
1047 | sizeBlock::value, |
1048 | typename SparseGridType::chunking_type> |
1049 | (miss_al_map,mp_off, |
1050 | b_src,b_dst,vk); |
1051 | |
1052 | // Now we intersect every chunk source |
1053 | |
1054 | Box<SparseGridType::dims,long int> b_inte; |
1055 | Box<SparseGridType::dims,long int> b_c; |
1056 | |
1057 | auto & data_src = grid_src.private_get_data(); |
1058 | auto & = grid_src.private_get_header(); |
1059 | |
1060 | auto & data_dst = grid_src.private_get_data(); |
1061 | auto & = grid_src.private_get_header(); |
1062 | |
1063 | grid_sm<SparseGridType::dims,void> gb; |
1064 | |
1065 | int chunk_pos[openfpm::math::pow(2,SparseGridType::dims)]; |
1066 | |
1067 | for (int i = 0 ; i < header_src.size() ; i++) |
1068 | { |
1069 | for (int j = 0 ; j < SparseGridType::dims ; j++) |
1070 | { |
1071 | b_c.setLow(j,header_src.get(i).pos.get(j)); |
1072 | b_c.setHigh(j,header_src.get(i).pos.get(j) + gbs.sz_block[j] - 1); |
1073 | } |
1074 | |
1075 | bool inte = b_src.Intersect(b_c,b_inte); |
1076 | |
1077 | if (inte == true) |
1078 | { |
1079 | // Prepare destination chunks |
1080 | |
1081 | for (int s = 0 ; s < vk.size() ; s++) |
1082 | { |
1083 | chunk_pos[vk.get(s).i] = grid_dst.getChunkCreate(); |
1084 | } |
1085 | |
1086 | grid_key_dx_iterator_sub<SparseGridType::dims> it(gb,b_inte.getKP1(),b_inte.getKP2()); |
1087 | |
1088 | auto & block_src = data_src.get(i); |
1089 | |
1090 | while (it.isNext()) |
1091 | { |
1092 | auto id = gb.LinId(it.get()); |
1093 | |
1094 | int dest_bid = chunk_pos[miss_al_map[id]]; |
1095 | |
1096 | data_dst.get(dest_bid)[mp_off[id]] = block_src[i]; |
1097 | |
1098 | ++it; |
1099 | } |
1100 | } |
1101 | } |
1102 | } |
1103 | |
1104 | #endif /* SPARSEGRID_CHUNK_COPY_HPP_ */ |
1105 | |