1 | /*************************************************************************** |
---|
2 | * blitz/array/stencils.cc Apply stencils to arrays |
---|
3 | * |
---|
4 | * $Id$ |
---|
5 | * |
---|
6 | * Copyright (C) 1997-2011 Todd Veldhuizen <tveldhui@acm.org> |
---|
7 | * |
---|
8 | * This file is a part of Blitz. |
---|
9 | * |
---|
10 | * Blitz is free software: you can redistribute it and/or modify |
---|
11 | * it under the terms of the GNU Lesser General Public License |
---|
12 | * as published by the Free Software Foundation, either version 3 |
---|
13 | * of the License, or (at your option) any later version. |
---|
14 | * |
---|
15 | * Blitz is distributed in the hope that it will be useful, |
---|
16 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
---|
17 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
---|
18 | * GNU Lesser General Public License for more details. |
---|
19 | * |
---|
20 | * You should have received a copy of the GNU Lesser General Public |
---|
21 | * License along with Blitz. If not, see <http://www.gnu.org/licenses/>. |
---|
22 | * |
---|
23 | * Suggestions: blitz-devel@lists.sourceforge.net |
---|
24 | * Bugs: blitz-support@lists.sourceforge.net |
---|
25 | * |
---|
26 | * For more information, please see the Blitz++ Home Page: |
---|
27 | * https://sourceforge.net/projects/blitz/ |
---|
28 | * |
---|
29 | ****************************************************************************/ |
---|
30 | #ifndef BZ_ARRAYSTENCILS_CC |
---|
31 | #define BZ_ARRAYSTENCILS_CC |
---|
32 | |
---|
33 | #ifndef BZ_ARRAYSTENCILS_H |
---|
34 | #error <blitz/array/stencil.cc> must be included via <blitz/array/stencils.h> |
---|
35 | #endif |
---|
36 | |
---|
37 | BZ_NAMESPACE(blitz) |
---|
38 | |
---|
39 | // NEEDS_WORK: |
---|
40 | // o Need to allow scalar arguments as well as arrays |
---|
41 | // o Unit stride optimization |
---|
42 | // o Tiling |
---|
43 | // o Pass coordinate vector to stencil, so that where-like constructs |
---|
44 | // can depend on location |
---|
45 | // o Maybe allow expression templates to be passed as |
---|
46 | // array parameters? |
---|
47 | |
---|
48 | /* |
---|
49 | * There are a lot of kludges in this code to work around the fact that |
---|
50 | * you can't have default template parameters with function templates. |
---|
51 | * Ideally, one would implement applyStencil(..) as: |
---|
52 | * |
---|
53 | * template<typename T_stencil, typename T_numtype1, typename T_array2, |
---|
54 | * class T_array3, typename T_array4, typename T_array5, typename T_array6, |
---|
55 | * class T_array7, typename T_array8, typename T_array9, typename T_array10, |
---|
56 | * class T_array11> |
---|
57 | * void applyStencil(const T_stencil& stencil, Array<T_numtype1,3>& A, |
---|
58 | * T_array2& B = _dummyArray, T_array3& C = _dummyArray, ......) |
---|
59 | * |
---|
60 | * and allow for up to (say) 11 arrays to be passed. But this doesn't |
---|
61 | * appear to be legal C++. Instead, 11 versions of applyStencil are |
---|
62 | * provided, each one with a different number of array parameters, |
---|
63 | * and these stubs fill in the _dummyArray parameters and invoke |
---|
64 | * applyStencil_imp(). |
---|
65 | */ |
---|
66 | |
---|
67 | template<int N_rank, typename T_numtype1, typename T_array2, |
---|
68 | class T_array3, typename T_array4, typename T_array5, typename T_array6, |
---|
69 | class T_array7, typename T_array8, typename T_array9, typename T_array10, |
---|
70 | class T_array11> |
---|
71 | void checkShapes(const Array<T_numtype1,N_rank>& BZ_DEBUG_PARAM(A), |
---|
72 | const T_array2& BZ_DEBUG_PARAM(B), const T_array3& BZ_DEBUG_PARAM(C), |
---|
73 | const T_array4& BZ_DEBUG_PARAM(D), const T_array5& BZ_DEBUG_PARAM(E), |
---|
74 | const T_array6& BZ_DEBUG_PARAM(F), const T_array7& BZ_DEBUG_PARAM(G), |
---|
75 | const T_array8& BZ_DEBUG_PARAM(H), const T_array9& BZ_DEBUG_PARAM(I), |
---|
76 | const T_array10& BZ_DEBUG_PARAM(J), const T_array11& BZ_DEBUG_PARAM(K)) |
---|
77 | { |
---|
78 | BZPRECONDITION(areShapesConformable(A.shape(),B.shape()) |
---|
79 | && areShapesConformable(A.shape(),C.shape()) |
---|
80 | && areShapesConformable(A.shape(),D.shape()) |
---|
81 | && areShapesConformable(A.shape(),E.shape()) |
---|
82 | && areShapesConformable(A.shape(),F.shape()) |
---|
83 | && areShapesConformable(A.shape(),G.shape()) |
---|
84 | && areShapesConformable(A.shape(),H.shape()) |
---|
85 | && areShapesConformable(A.shape(),I.shape()) |
---|
86 | && areShapesConformable(A.shape(),J.shape()) |
---|
87 | && areShapesConformable(A.shape(),K.shape())); |
---|
88 | } |
---|
89 | |
---|
90 | template<typename T_extent, int N_rank, |
---|
91 | class T_stencil, typename T_numtype1, typename T_array2, |
---|
92 | class T_array3, typename T_array4, typename T_array5, typename T_array6, |
---|
93 | class T_array7, typename T_array8, typename T_array9, typename T_array10, |
---|
94 | class T_array11> |
---|
95 | void calcStencilExtent(T_extent& At, const T_stencil& stencil, |
---|
96 | const Array<T_numtype1,N_rank>&, |
---|
97 | const T_array2&, const T_array3&, const T_array4&, const T_array5&, |
---|
98 | const T_array6&, const T_array7&, const T_array8&, const T_array9&, |
---|
99 | const T_array10&, const T_array11&) |
---|
100 | { |
---|
101 | // Interrogate the stencil to find out its extent |
---|
102 | _bz_typename stencilExtent_traits<T_array2>::T_stencilExtent Bt; |
---|
103 | _bz_typename stencilExtent_traits<T_array3>::T_stencilExtent Ct; |
---|
104 | _bz_typename stencilExtent_traits<T_array4>::T_stencilExtent Dt; |
---|
105 | _bz_typename stencilExtent_traits<T_array5>::T_stencilExtent Et; |
---|
106 | _bz_typename stencilExtent_traits<T_array6>::T_stencilExtent Ft; |
---|
107 | _bz_typename stencilExtent_traits<T_array7>::T_stencilExtent Gt; |
---|
108 | _bz_typename stencilExtent_traits<T_array8>::T_stencilExtent Ht; |
---|
109 | _bz_typename stencilExtent_traits<T_array9>::T_stencilExtent It; |
---|
110 | _bz_typename stencilExtent_traits<T_array10>::T_stencilExtent Jt; |
---|
111 | _bz_typename stencilExtent_traits<T_array11>::T_stencilExtent Kt; |
---|
112 | |
---|
113 | stencil.apply(At, Bt, Ct, Dt, Et, Ft, Gt, Ht, It, Jt, Kt); |
---|
114 | At.combine(Bt); |
---|
115 | At.combine(Ct); |
---|
116 | At.combine(Dt); |
---|
117 | At.combine(Et); |
---|
118 | At.combine(Ft); |
---|
119 | At.combine(Gt); |
---|
120 | At.combine(Ht); |
---|
121 | At.combine(It); |
---|
122 | At.combine(Jt); |
---|
123 | At.combine(Kt); |
---|
124 | } |
---|
125 | |
---|
126 | template<int N_rank, typename T_stencil, typename T_numtype1, typename T_array2> |
---|
127 | RectDomain<N_rank> interiorDomain(const T_stencil& stencil, |
---|
128 | const Array<T_numtype1,N_rank>& A, |
---|
129 | const T_array2& B) |
---|
130 | { |
---|
131 | RectDomain<N_rank> domain = A.domain(); |
---|
132 | |
---|
133 | // Interrogate the stencil to find out its extent |
---|
134 | stencilExtent<3, T_numtype1> At; |
---|
135 | calcStencilExtent(At, stencil, A, B, _dummyArray, _dummyArray, |
---|
136 | _dummyArray, _dummyArray, _dummyArray, _dummyArray, _dummyArray, |
---|
137 | _dummyArray, _dummyArray); |
---|
138 | |
---|
139 | // Shrink the domain according to the stencil size |
---|
140 | TinyVector<int,N_rank> lbound, ubound; |
---|
141 | lbound = domain.lbound() - (At.min)(); |
---|
142 | ubound = domain.ubound() - (At.max)(); |
---|
143 | return RectDomain<N_rank>(lbound,ubound); |
---|
144 | } |
---|
145 | |
---|
146 | template<int hasExtents> |
---|
147 | struct _getStencilExtent { |
---|
148 | template<int N_rank, |
---|
149 | class T_stencil, typename T_numtype1, typename T_array2, |
---|
150 | class T_array3, typename T_array4, typename T_array5, typename T_array6, |
---|
151 | class T_array7, typename T_array8, typename T_array9, typename T_array10, |
---|
152 | class T_array11> |
---|
153 | static void getStencilExtent(TinyVector<int,N_rank>& minb, |
---|
154 | TinyVector<int,N_rank>& maxb, |
---|
155 | const T_stencil& stencil, Array<T_numtype1,N_rank>& A, |
---|
156 | T_array2& B, T_array3& C, T_array4& D, T_array5& E, T_array6& F, |
---|
157 | T_array7& G, T_array8& H, T_array9& I, T_array10& J, T_array11& K) |
---|
158 | { |
---|
159 | // Interrogate the stencil to find out its extent |
---|
160 | stencilExtent<N_rank, T_numtype1> At; |
---|
161 | calcStencilExtent(At, stencil, A, B, C, D, E, F, G, H, I, J, K); |
---|
162 | minb = (At.min)(); |
---|
163 | maxb = (At.max)(); |
---|
164 | } |
---|
165 | }; |
---|
166 | |
---|
167 | template<> |
---|
168 | struct _getStencilExtent<1> { |
---|
169 | template<int N_rank, |
---|
170 | class T_stencil, typename T_numtype1, typename T_array2, |
---|
171 | class T_array3, typename T_array4, typename T_array5, typename T_array6, |
---|
172 | class T_array7, typename T_array8, typename T_array9, typename T_array10, |
---|
173 | class T_array11> |
---|
174 | static inline void getStencilExtent(TinyVector<int,N_rank>& minb, |
---|
175 | TinyVector<int,N_rank>& maxb, |
---|
176 | const T_stencil& stencil, Array<T_numtype1,N_rank>&, |
---|
177 | T_array2&, T_array3&, T_array4&, T_array5&, T_array6&, |
---|
178 | T_array7&, T_array8&, T_array9&, T_array10&, T_array11&) |
---|
179 | { |
---|
180 | stencil.getExtent(minb, maxb); |
---|
181 | } |
---|
182 | }; |
---|
183 | |
---|
184 | template<int N_rank, |
---|
185 | class T_stencil, typename T_numtype1, typename T_array2, |
---|
186 | class T_array3, typename T_array4, typename T_array5, typename T_array6, |
---|
187 | class T_array7, typename T_array8, typename T_array9, typename T_array10, |
---|
188 | class T_array11> |
---|
189 | inline void getStencilExtent(TinyVector<int,N_rank>& minb, |
---|
190 | TinyVector<int,N_rank>& maxb, |
---|
191 | const T_stencil& stencil, Array<T_numtype1,N_rank>& A, |
---|
192 | T_array2& B, T_array3& C, T_array4& D, T_array5& E, T_array6& F, |
---|
193 | T_array7& G, T_array8& H, T_array9& I, T_array10& J, T_array11& K) |
---|
194 | { |
---|
195 | _getStencilExtent<T_stencil::hasExtent>::getStencilExtent( |
---|
196 | minb, maxb, stencil, A, B, C, D, E, F, G, H, I, J, K); |
---|
197 | } |
---|
198 | |
---|
199 | /* |
---|
200 | * This version applies a stencil to a set of 3D arrays. Up to 11 arrays |
---|
201 | * may be used. Any unused arrays are turned into dummyArray objects. |
---|
202 | * Operations on dummyArray objects are translated into no-ops. |
---|
203 | */ |
---|
204 | template<typename T_stencil, typename T_numtype1, typename T_array2, |
---|
205 | class T_array3, typename T_array4, typename T_array5, typename T_array6, |
---|
206 | class T_array7, typename T_array8, typename T_array9, typename T_array10, |
---|
207 | class T_array11> |
---|
208 | void applyStencil_imp(const T_stencil& stencil, Array<T_numtype1,3>& A, |
---|
209 | T_array2& B, T_array3& C, T_array4& D, T_array5& E, T_array6& F, |
---|
210 | T_array7& G, T_array8& H, T_array9& I, T_array10& J, T_array11& K) |
---|
211 | { |
---|
212 | checkShapes(A,B,C,D,E,F,G,H,I,J,K); |
---|
213 | |
---|
214 | // Determine stencil extent |
---|
215 | TinyVector<int,3> minb, maxb; |
---|
216 | getStencilExtent(minb, maxb, stencil, A, B, C, D, E, F, G, H, I, J, K); |
---|
217 | |
---|
218 | // Now determine the subdomain over which the stencil |
---|
219 | // can be applied without worrying about overrunning the |
---|
220 | // boundaries of the array |
---|
221 | int stencil_lbound0 = minb(0); |
---|
222 | int stencil_lbound1 = minb(1); |
---|
223 | int stencil_lbound2 = minb(2); |
---|
224 | |
---|
225 | int stencil_ubound0 = maxb(0); |
---|
226 | int stencil_ubound1 = maxb(1); |
---|
227 | int stencil_ubound2 = maxb(2); |
---|
228 | |
---|
229 | int lbound0 = (extrema::max)(A.lbound(0), A.lbound(0) - stencil_lbound0); |
---|
230 | int lbound1 = (extrema::max)(A.lbound(1), A.lbound(1) - stencil_lbound1); |
---|
231 | int lbound2 = (extrema::max)(A.lbound(2), A.lbound(2) - stencil_lbound2); |
---|
232 | |
---|
233 | int ubound0 = (extrema::min)(A.ubound(0), A.ubound(0) - stencil_ubound0); |
---|
234 | int ubound1 = (extrema::min)(A.ubound(1), A.ubound(1) - stencil_ubound1); |
---|
235 | int ubound2 = (extrema::min)(A.ubound(2), A.ubound(2) - stencil_ubound2); |
---|
236 | |
---|
237 | #if 0 |
---|
238 | cout << "Stencil bounds are:" << endl |
---|
239 | << lbound0 << '\t' << ubound0 << endl |
---|
240 | << lbound1 << '\t' << ubound1 << endl |
---|
241 | << lbound2 << '\t' << ubound2 << endl; |
---|
242 | #endif |
---|
243 | |
---|
244 | // Now do the actual loop |
---|
245 | FastArrayIterator<T_numtype1,3> Aiter(A); |
---|
246 | _bz_typename T_array2::T_iterator Biter(B); |
---|
247 | _bz_typename T_array3::T_iterator Citer(C); |
---|
248 | _bz_typename T_array4::T_iterator Diter(D); |
---|
249 | _bz_typename T_array5::T_iterator Eiter(E); |
---|
250 | _bz_typename T_array6::T_iterator Fiter(F); |
---|
251 | _bz_typename T_array7::T_iterator Giter(G); |
---|
252 | _bz_typename T_array8::T_iterator Hiter(H); |
---|
253 | _bz_typename T_array9::T_iterator Iiter(I); |
---|
254 | _bz_typename T_array10::T_iterator Jiter(J); |
---|
255 | _bz_typename T_array11::T_iterator Kiter(K); |
---|
256 | |
---|
257 | // Load the strides for the innermost loop |
---|
258 | Aiter.loadStride(2); |
---|
259 | Biter.loadStride(2); |
---|
260 | Citer.loadStride(2); |
---|
261 | Diter.loadStride(2); |
---|
262 | Eiter.loadStride(2); |
---|
263 | Fiter.loadStride(2); |
---|
264 | Giter.loadStride(2); |
---|
265 | Hiter.loadStride(2); |
---|
266 | Iiter.loadStride(2); |
---|
267 | Jiter.loadStride(2); |
---|
268 | Kiter.loadStride(2); |
---|
269 | |
---|
270 | for (int i=lbound0; i <= ubound0; ++i) |
---|
271 | { |
---|
272 | for (int j=lbound1; j <= ubound1; ++j) |
---|
273 | { |
---|
274 | Aiter.moveTo(i,j,lbound2); |
---|
275 | Biter.moveTo(i,j,lbound2); |
---|
276 | Citer.moveTo(i,j,lbound2); |
---|
277 | Diter.moveTo(i,j,lbound2); |
---|
278 | Eiter.moveTo(i,j,lbound2); |
---|
279 | Fiter.moveTo(i,j,lbound2); |
---|
280 | Giter.moveTo(i,j,lbound2); |
---|
281 | Hiter.moveTo(i,j,lbound2); |
---|
282 | Iiter.moveTo(i,j,lbound2); |
---|
283 | Jiter.moveTo(i,j,lbound2); |
---|
284 | Kiter.moveTo(i,j,lbound2); |
---|
285 | |
---|
286 | for (int k=lbound2; k <= ubound2; ++k) |
---|
287 | { |
---|
288 | stencil.apply(Aiter, Biter, Citer, Diter, Eiter, Fiter, Giter, |
---|
289 | Hiter, Iiter, Jiter, Kiter); |
---|
290 | |
---|
291 | Aiter.advance(); |
---|
292 | Biter.advance(); |
---|
293 | Citer.advance(); |
---|
294 | Diter.advance(); |
---|
295 | Eiter.advance(); |
---|
296 | Fiter.advance(); |
---|
297 | Giter.advance(); |
---|
298 | Hiter.advance(); |
---|
299 | Iiter.advance(); |
---|
300 | Jiter.advance(); |
---|
301 | Kiter.advance(); |
---|
302 | } |
---|
303 | } |
---|
304 | } |
---|
305 | } |
---|
306 | |
---|
307 | /* |
---|
308 | * This version applies a stencil to a set of 2D arrays. Up to 11 arrays |
---|
309 | * may be used. Any unused arrays are turned into dummyArray objects. |
---|
310 | * Operations on dummyArray objects are translated into no-ops. |
---|
311 | */ |
---|
312 | template<typename T_stencil, typename T_numtype1, typename T_array2, |
---|
313 | class T_array3, typename T_array4, typename T_array5, typename T_array6, |
---|
314 | class T_array7, typename T_array8, typename T_array9, typename T_array10, |
---|
315 | class T_array11> |
---|
316 | void applyStencil_imp(const T_stencil& stencil, Array<T_numtype1,2>& A, |
---|
317 | T_array2& B, T_array3& C, T_array4& D, T_array5& E, T_array6& F, |
---|
318 | T_array7& G, T_array8& H, T_array9& I, T_array10& J, T_array11& K) |
---|
319 | { |
---|
320 | checkShapes(A,B,C,D,E,F,G,H,I,J,K); |
---|
321 | |
---|
322 | // Determine stencil extent |
---|
323 | TinyVector<int,2> minb, maxb; |
---|
324 | getStencilExtent(minb, maxb, stencil, A, B, C, D, E, F, G, H, I, J, K); |
---|
325 | |
---|
326 | // Now determine the subdomain over which the stencil |
---|
327 | // can be applied without worrying about overrunning the |
---|
328 | // boundaries of the array |
---|
329 | int stencil_lbound0 = minb(0); |
---|
330 | int stencil_lbound1 = minb(1); |
---|
331 | |
---|
332 | int stencil_ubound0 = maxb(0); |
---|
333 | int stencil_ubound1 = maxb(1); |
---|
334 | |
---|
335 | int lbound0 = (extrema::max)(A.lbound(0), A.lbound(0) - stencil_lbound0); |
---|
336 | int lbound1 = (extrema::max)(A.lbound(1), A.lbound(1) - stencil_lbound1); |
---|
337 | |
---|
338 | int ubound0 = (extrema::min)(A.ubound(0), A.ubound(0) - stencil_ubound0); |
---|
339 | int ubound1 = (extrema::min)(A.ubound(1), A.ubound(1) - stencil_ubound1); |
---|
340 | |
---|
341 | #if 0 |
---|
342 | cout << "Stencil bounds are:" << endl |
---|
343 | << lbound0 << '\t' << ubound0 << endl |
---|
344 | << lbound1 << '\t' << ubound1 << endl; |
---|
345 | #endif |
---|
346 | |
---|
347 | // Now do the actual loop |
---|
348 | FastArrayIterator<T_numtype1,2> Aiter(A); |
---|
349 | _bz_typename T_array2::T_iterator Biter(B); |
---|
350 | _bz_typename T_array3::T_iterator Citer(C); |
---|
351 | _bz_typename T_array4::T_iterator Diter(D); |
---|
352 | _bz_typename T_array5::T_iterator Eiter(E); |
---|
353 | _bz_typename T_array6::T_iterator Fiter(F); |
---|
354 | _bz_typename T_array7::T_iterator Giter(G); |
---|
355 | _bz_typename T_array8::T_iterator Hiter(H); |
---|
356 | _bz_typename T_array9::T_iterator Iiter(I); |
---|
357 | _bz_typename T_array10::T_iterator Jiter(J); |
---|
358 | _bz_typename T_array11::T_iterator Kiter(K); |
---|
359 | |
---|
360 | // Load the strides for the innermost loop |
---|
361 | Aiter.loadStride(1); |
---|
362 | Biter.loadStride(1); |
---|
363 | Citer.loadStride(1); |
---|
364 | Diter.loadStride(1); |
---|
365 | Eiter.loadStride(1); |
---|
366 | Fiter.loadStride(1); |
---|
367 | Giter.loadStride(1); |
---|
368 | Hiter.loadStride(1); |
---|
369 | Iiter.loadStride(1); |
---|
370 | Jiter.loadStride(1); |
---|
371 | Kiter.loadStride(1); |
---|
372 | |
---|
373 | for (int i=lbound0; i <= ubound0; ++i) |
---|
374 | { |
---|
375 | Aiter.moveTo(i,lbound1); |
---|
376 | Biter.moveTo(i,lbound1); |
---|
377 | Citer.moveTo(i,lbound1); |
---|
378 | Diter.moveTo(i,lbound1); |
---|
379 | Eiter.moveTo(i,lbound1); |
---|
380 | Fiter.moveTo(i,lbound1); |
---|
381 | Giter.moveTo(i,lbound1); |
---|
382 | Hiter.moveTo(i,lbound1); |
---|
383 | Iiter.moveTo(i,lbound1); |
---|
384 | Jiter.moveTo(i,lbound1); |
---|
385 | Kiter.moveTo(i,lbound1); |
---|
386 | |
---|
387 | for (int k=lbound1; k <= ubound1; ++k) |
---|
388 | { |
---|
389 | stencil.apply(Aiter, Biter, Citer, Diter, Eiter, Fiter, Giter, |
---|
390 | Hiter, Iiter, Jiter, Kiter); |
---|
391 | |
---|
392 | Aiter.advance(); |
---|
393 | Biter.advance(); |
---|
394 | Citer.advance(); |
---|
395 | Diter.advance(); |
---|
396 | Eiter.advance(); |
---|
397 | Fiter.advance(); |
---|
398 | Giter.advance(); |
---|
399 | Hiter.advance(); |
---|
400 | Iiter.advance(); |
---|
401 | Jiter.advance(); |
---|
402 | Kiter.advance(); |
---|
403 | } |
---|
404 | } |
---|
405 | } |
---|
406 | |
---|
407 | /* |
---|
408 | * This version applies a stencil to a set of 1D arrays. Up to 11 arrays |
---|
409 | * may be used. Any unused arrays are turned into dummyArray objects. |
---|
410 | * Operations on dummyArray objects are translated into no-ops. |
---|
411 | */ |
---|
412 | template<typename T_stencil, typename T_numtype1, typename T_array2, |
---|
413 | class T_array3, typename T_array4, typename T_array5, typename T_array6, |
---|
414 | class T_array7, typename T_array8, typename T_array9, typename T_array10, |
---|
415 | class T_array11> |
---|
416 | void applyStencil_imp(const T_stencil& stencil, Array<T_numtype1,1>& A, |
---|
417 | T_array2& B, T_array3& C, T_array4& D, T_array5& E, T_array6& F, |
---|
418 | T_array7& G, T_array8& H, T_array9& I, T_array10& J, T_array11& K) |
---|
419 | { |
---|
420 | checkShapes(A,B,C,D,E,F,G,H,I,J,K); |
---|
421 | |
---|
422 | // Determine stencil extent |
---|
423 | TinyVector<int,1> minb, maxb; |
---|
424 | getStencilExtent(minb, maxb, stencil, A, B, C, D, E, F, G, H, I, J, K); |
---|
425 | |
---|
426 | // Now determine the subdomain over which the stencil |
---|
427 | // can be applied without worrying about overrunning the |
---|
428 | // boundaries of the array |
---|
429 | int stencil_lbound0 = minb(0); |
---|
430 | int stencil_ubound0 = maxb(0); |
---|
431 | |
---|
432 | int lbound0 = (extrema::max)(A.lbound(0), A.lbound(0) - stencil_lbound0); |
---|
433 | int ubound0 = (extrema::min)(A.ubound(0), A.ubound(0) - stencil_ubound0); |
---|
434 | |
---|
435 | #if 0 |
---|
436 | cout << "Stencil bounds are:" << endl |
---|
437 | << lbound0 << '\t' << ubound0 << endl; |
---|
438 | #endif |
---|
439 | |
---|
440 | // Now do the actual loop |
---|
441 | FastArrayIterator<T_numtype1,1> Aiter(A); |
---|
442 | _bz_typename T_array2::T_iterator Biter(B); |
---|
443 | _bz_typename T_array3::T_iterator Citer(C); |
---|
444 | _bz_typename T_array4::T_iterator Diter(D); |
---|
445 | _bz_typename T_array5::T_iterator Eiter(E); |
---|
446 | _bz_typename T_array6::T_iterator Fiter(F); |
---|
447 | _bz_typename T_array7::T_iterator Giter(G); |
---|
448 | _bz_typename T_array8::T_iterator Hiter(H); |
---|
449 | _bz_typename T_array9::T_iterator Iiter(I); |
---|
450 | _bz_typename T_array10::T_iterator Jiter(J); |
---|
451 | _bz_typename T_array11::T_iterator Kiter(K); |
---|
452 | |
---|
453 | // Load the strides for the innermost loop |
---|
454 | Aiter.loadStride(0); |
---|
455 | Biter.loadStride(0); |
---|
456 | Citer.loadStride(0); |
---|
457 | Diter.loadStride(0); |
---|
458 | Eiter.loadStride(0); |
---|
459 | Fiter.loadStride(0); |
---|
460 | Giter.loadStride(0); |
---|
461 | Hiter.loadStride(0); |
---|
462 | Iiter.loadStride(0); |
---|
463 | Jiter.loadStride(0); |
---|
464 | Kiter.loadStride(0); |
---|
465 | |
---|
466 | // Patch from Derrick Bass |
---|
467 | Aiter.moveTo(lbound0); |
---|
468 | Biter.moveTo(lbound0); |
---|
469 | Citer.moveTo(lbound0); |
---|
470 | Diter.moveTo(lbound0); |
---|
471 | Eiter.moveTo(lbound0); |
---|
472 | Fiter.moveTo(lbound0); |
---|
473 | Giter.moveTo(lbound0); |
---|
474 | Hiter.moveTo(lbound0); |
---|
475 | Iiter.moveTo(lbound0); |
---|
476 | Jiter.moveTo(lbound0); |
---|
477 | Kiter.moveTo(lbound0); |
---|
478 | |
---|
479 | for (int i=lbound0; i <= ubound0; ++i) |
---|
480 | { |
---|
481 | stencil.apply(Aiter, Biter, Citer, Diter, Eiter, Fiter, Giter, |
---|
482 | Hiter, Iiter, Jiter, Kiter); |
---|
483 | |
---|
484 | Aiter.advance(); |
---|
485 | Biter.advance(); |
---|
486 | Citer.advance(); |
---|
487 | Diter.advance(); |
---|
488 | Eiter.advance(); |
---|
489 | Fiter.advance(); |
---|
490 | Giter.advance(); |
---|
491 | Hiter.advance(); |
---|
492 | Iiter.advance(); |
---|
493 | Jiter.advance(); |
---|
494 | Kiter.advance(); |
---|
495 | } |
---|
496 | } |
---|
497 | |
---|
498 | /* |
---|
499 | * These 11 versions of applyStencil handle from 1 to 11 array parameters. |
---|
500 | * They pad their argument list with enough dummyArray objects to call |
---|
501 | * applyStencil_imp with 11 array parameters. |
---|
502 | */ |
---|
503 | template<typename T_stencil, typename T_numtype1, int N_rank> |
---|
504 | inline void applyStencil(const T_stencil& stencil, Array<T_numtype1,N_rank>& A) |
---|
505 | { |
---|
506 | applyStencil_imp(stencil, A, _dummyArray, _dummyArray, |
---|
507 | _dummyArray, _dummyArray, _dummyArray, _dummyArray, |
---|
508 | _dummyArray, _dummyArray, _dummyArray, _dummyArray); |
---|
509 | } |
---|
510 | |
---|
511 | template<typename T_stencil, typename T_numtype1, int N_rank, typename T_array2> |
---|
512 | inline void applyStencil(const T_stencil& stencil, Array<T_numtype1,N_rank>& A, |
---|
513 | T_array2& B) |
---|
514 | { |
---|
515 | applyStencil_imp(stencil, A, B, _dummyArray, _dummyArray, |
---|
516 | _dummyArray, _dummyArray, _dummyArray, _dummyArray, |
---|
517 | _dummyArray, _dummyArray, _dummyArray); |
---|
518 | } |
---|
519 | |
---|
520 | template<typename T_stencil, typename T_numtype1, int N_rank, typename T_array2, |
---|
521 | class T_array3> |
---|
522 | inline void applyStencil(const T_stencil& stencil, Array<T_numtype1,N_rank>& A, |
---|
523 | T_array2& B, T_array3& C) |
---|
524 | { |
---|
525 | applyStencil_imp(stencil, A, B, C, _dummyArray, _dummyArray, |
---|
526 | _dummyArray, _dummyArray, _dummyArray, _dummyArray, _dummyArray, |
---|
527 | _dummyArray); |
---|
528 | } |
---|
529 | |
---|
530 | template<typename T_stencil, typename T_numtype1, int N_rank, typename T_array2, |
---|
531 | class T_array3, typename T_array4> |
---|
532 | inline void applyStencil(const T_stencil& stencil, Array<T_numtype1,N_rank>& A, |
---|
533 | T_array2& B, T_array3& C, T_array4& D) |
---|
534 | { |
---|
535 | applyStencil_imp(stencil, A, B, C, D, _dummyArray, _dummyArray, |
---|
536 | _dummyArray, _dummyArray, _dummyArray, _dummyArray, _dummyArray); |
---|
537 | } |
---|
538 | |
---|
539 | template<typename T_stencil, typename T_numtype1, int N_rank, typename T_array2, |
---|
540 | class T_array3, typename T_array4, typename T_array5> |
---|
541 | inline void applyStencil(const T_stencil& stencil, Array<T_numtype1,N_rank>& A, |
---|
542 | T_array2& B, T_array3& C, T_array4& D, T_array5& E) |
---|
543 | { |
---|
544 | applyStencil_imp(stencil, A, B, C, D, E, _dummyArray, |
---|
545 | _dummyArray, _dummyArray, _dummyArray, _dummyArray, _dummyArray); |
---|
546 | } |
---|
547 | |
---|
548 | template<typename T_stencil, typename T_numtype1, int N_rank, typename T_array2, |
---|
549 | class T_array3, typename T_array4, typename T_array5, typename T_array6> |
---|
550 | inline void applyStencil(const T_stencil& stencil, Array<T_numtype1,N_rank>& A, |
---|
551 | T_array2& B, T_array3& C, T_array4& D, T_array5& E, T_array6& F) |
---|
552 | { |
---|
553 | applyStencil_imp(stencil, A, B, C, D, E, F, |
---|
554 | _dummyArray, _dummyArray, _dummyArray, _dummyArray, _dummyArray); |
---|
555 | } |
---|
556 | |
---|
557 | template<typename T_stencil, typename T_numtype1, int N_rank, typename T_array2, |
---|
558 | class T_array3, typename T_array4, typename T_array5, typename T_array6, |
---|
559 | class T_array7> |
---|
560 | inline void applyStencil(const T_stencil& stencil, Array<T_numtype1,N_rank>& A, |
---|
561 | T_array2& B, T_array3& C, T_array4& D, T_array5& E, T_array6& F, |
---|
562 | T_array7& G) |
---|
563 | { |
---|
564 | applyStencil_imp(stencil, A, B, C, D, E, F, G, |
---|
565 | _dummyArray, _dummyArray, _dummyArray, _dummyArray); |
---|
566 | } |
---|
567 | |
---|
568 | template<typename T_stencil, typename T_numtype1, int N_rank, typename T_array2, |
---|
569 | class T_array3, typename T_array4, typename T_array5, typename T_array6, |
---|
570 | class T_array7, typename T_array8> |
---|
571 | inline void applyStencil(const T_stencil& stencil, Array<T_numtype1,N_rank>& A, |
---|
572 | T_array2& B, T_array3& C, T_array4& D, T_array5& E, T_array6& F, |
---|
573 | T_array7& G, T_array8& H) |
---|
574 | { |
---|
575 | applyStencil_imp(stencil, A, B, C, D, E, F, G, H, |
---|
576 | _dummyArray, _dummyArray, _dummyArray); |
---|
577 | } |
---|
578 | |
---|
579 | template<typename T_stencil, typename T_numtype1, int N_rank, typename T_array2, |
---|
580 | class T_array3, typename T_array4, typename T_array5, typename T_array6, |
---|
581 | class T_array7, typename T_array8, typename T_array9> |
---|
582 | inline void applyStencil(const T_stencil& stencil, Array<T_numtype1,N_rank>& A, |
---|
583 | T_array2& B, T_array3& C, T_array4& D, T_array5& E, T_array6& F, |
---|
584 | T_array7& G, T_array8& H, T_array9& I) |
---|
585 | { |
---|
586 | applyStencil_imp(stencil, A, B, C, D, E, F, G, H, I, |
---|
587 | _dummyArray, _dummyArray); |
---|
588 | } |
---|
589 | |
---|
590 | template<typename T_stencil, typename T_numtype1, int N_rank, typename T_array2, |
---|
591 | class T_array3, typename T_array4, typename T_array5, typename T_array6, |
---|
592 | class T_array7, typename T_array8, typename T_array9, typename T_array10> |
---|
593 | inline void applyStencil(const T_stencil& stencil, Array<T_numtype1,N_rank>& A, |
---|
594 | T_array2& B, T_array3& C, T_array4& D, T_array5& E, T_array6& F, |
---|
595 | T_array7& G, T_array8& H, T_array9& I, T_array10& J) |
---|
596 | { |
---|
597 | applyStencil_imp(stencil, A, B, C, D, E, F, G, H, I, J, |
---|
598 | _dummyArray); |
---|
599 | } |
---|
600 | |
---|
601 | template<typename T_stencil, typename T_numtype1, int N_rank, typename T_array2, |
---|
602 | class T_array3, typename T_array4, typename T_array5, typename T_array6, |
---|
603 | class T_array7, typename T_array8, typename T_array9, typename T_array10, |
---|
604 | class T_array11> |
---|
605 | inline void applyStencil(const T_stencil& stencil, Array<T_numtype1,N_rank>& A, |
---|
606 | T_array2& B, T_array3& C, T_array4& D, T_array5& E, T_array6& F, |
---|
607 | T_array7& G, T_array8& H, T_array9& I, T_array10& J, T_array11& K) |
---|
608 | { |
---|
609 | applyStencil_imp(stencil, A, B, C, D, E, F, G, H, I, J, K); |
---|
610 | } |
---|
611 | |
---|
612 | BZ_NAMESPACE_END |
---|
613 | |
---|
614 | #endif // BZ_ARRAYSTENCIL_CC |
---|