/src/gdal/alg/gdalgrid.cpp
Line | Count | Source (jump to first uncovered line) |
1 | | /****************************************************************************** |
2 | | * |
3 | | * Project: GDAL Gridding API. |
4 | | * Purpose: Implementation of GDAL scattered data gridder. |
5 | | * Author: Andrey Kiselev, dron@ak4719.spb.edu |
6 | | * |
7 | | ****************************************************************************** |
8 | | * Copyright (c) 2007, Andrey Kiselev <dron@ak4719.spb.edu> |
9 | | * Copyright (c) 2009-2013, Even Rouault <even dot rouault at spatialys.com> |
10 | | * |
11 | | * SPDX-License-Identifier: MIT |
12 | | ****************************************************************************/ |
13 | | |
14 | | #include "cpl_port.h" |
15 | | #include "gdalgrid.h" |
16 | | #include "gdalgrid_priv.h" |
17 | | |
18 | | #include <cfloat> |
19 | | #include <climits> |
20 | | #include <cmath> |
21 | | #include <cstdlib> |
22 | | #include <cstring> |
23 | | |
24 | | #include <limits> |
25 | | #include <map> |
26 | | #include <utility> |
27 | | #include <algorithm> |
28 | | |
29 | | #include "cpl_conv.h" |
30 | | #include "cpl_cpu_features.h" |
31 | | #include "cpl_error.h" |
32 | | #include "cpl_multiproc.h" |
33 | | #include "cpl_progress.h" |
34 | | #include "cpl_quad_tree.h" |
35 | | #include "cpl_string.h" |
36 | | #include "cpl_vsi.h" |
37 | | #include "cpl_worker_thread_pool.h" |
38 | | #include "gdal.h" |
39 | | |
40 | | constexpr double TO_RADIANS = M_PI / 180.0; |
41 | | |
42 | | /************************************************************************/ |
43 | | /* GDALGridGetPointBounds() */ |
44 | | /************************************************************************/ |
45 | | |
46 | | static void GDALGridGetPointBounds(const void *hFeature, CPLRectObj *pBounds) |
47 | 0 | { |
48 | 0 | const GDALGridPoint *psPoint = static_cast<const GDALGridPoint *>(hFeature); |
49 | 0 | GDALGridXYArrays *psXYArrays = psPoint->psXYArrays; |
50 | 0 | const int i = psPoint->i; |
51 | 0 | const double dfX = psXYArrays->padfX[i]; |
52 | 0 | const double dfY = psXYArrays->padfY[i]; |
53 | 0 | pBounds->minx = dfX; |
54 | 0 | pBounds->miny = dfY; |
55 | 0 | pBounds->maxx = dfX; |
56 | 0 | pBounds->maxy = dfY; |
57 | 0 | } |
58 | | |
59 | | /************************************************************************/ |
60 | | /* GDALGridInverseDistanceToAPower() */ |
61 | | /************************************************************************/ |
62 | | |
63 | | /** |
64 | | * Inverse distance to a power. |
65 | | * |
66 | | * The Inverse Distance to a Power gridding method is a weighted average |
67 | | * interpolator. You should supply the input arrays with the scattered data |
68 | | * values including coordinates of every data point and output grid geometry. |
69 | | * The function will compute interpolated value for the given position in |
70 | | * output grid. |
71 | | * |
72 | | * For every grid node the resulting value \f$Z\f$ will be calculated using |
73 | | * formula: |
74 | | * |
75 | | * \f[ |
76 | | * Z=\frac{\sum_{i=1}^n{\frac{Z_i}{r_i^p}}}{\sum_{i=1}^n{\frac{1}{r_i^p}}} |
77 | | * \f] |
78 | | * |
79 | | * where |
80 | | * <ul> |
81 | | * <li> \f$Z_i\f$ is a known value at point \f$i\f$, |
82 | | * <li> \f$r_i\f$ is an Euclidean distance from the grid node |
83 | | * to point \f$i\f$, |
84 | | * <li> \f$p\f$ is a weighting power, |
85 | | * <li> \f$n\f$ is a total number of points in search ellipse. |
86 | | * </ul> |
87 | | * |
88 | | * In this method the weighting factor \f$w\f$ is |
89 | | * |
90 | | * \f[ |
91 | | * w=\frac{1}{r^p} |
92 | | * \f] |
93 | | * |
94 | | * @param poOptionsIn Algorithm parameters. This should point to |
95 | | * GDALGridInverseDistanceToAPowerOptions object. |
96 | | * @param nPoints Number of elements in input arrays. |
97 | | * @param padfX Input array of X coordinates. |
98 | | * @param padfY Input array of Y coordinates. |
99 | | * @param padfZ Input array of Z values. |
100 | | * @param dfXPoint X coordinate of the point to compute. |
101 | | * @param dfYPoint Y coordinate of the point to compute. |
102 | | * @param pdfValue Pointer to variable where the computed grid node value |
103 | | * will be returned. |
104 | | * @param hExtraParamsIn extra parameters (unused) |
105 | | * |
106 | | * @return CE_None on success or CE_Failure if something goes wrong. |
107 | | */ |
108 | | |
109 | | CPLErr GDALGridInverseDistanceToAPower(const void *poOptionsIn, GUInt32 nPoints, |
110 | | const double *padfX, const double *padfY, |
111 | | const double *padfZ, double dfXPoint, |
112 | | double dfYPoint, double *pdfValue, |
113 | | CPL_UNUSED void *hExtraParamsIn) |
114 | 0 | { |
115 | | // TODO: For optimization purposes pre-computed parameters should be moved |
116 | | // out of this routine to the calling function. |
117 | |
|
118 | 0 | const GDALGridInverseDistanceToAPowerOptions *const poOptions = |
119 | 0 | static_cast<const GDALGridInverseDistanceToAPowerOptions *>( |
120 | 0 | poOptionsIn); |
121 | | |
122 | | // Pre-compute search ellipse parameters. |
123 | 0 | const double dfRadius1Square = poOptions->dfRadius1 * poOptions->dfRadius1; |
124 | 0 | const double dfRadius2Square = poOptions->dfRadius2 * poOptions->dfRadius2; |
125 | 0 | const double dfR12Square = dfRadius1Square * dfRadius2Square; |
126 | | |
127 | | // Compute coefficients for coordinate system rotation. |
128 | 0 | const double dfAngle = TO_RADIANS * poOptions->dfAngle; |
129 | 0 | const bool bRotated = dfAngle != 0.0; |
130 | 0 | const double dfCoeff1 = bRotated ? cos(dfAngle) : 0.0; |
131 | 0 | const double dfCoeff2 = bRotated ? sin(dfAngle) : 0.0; |
132 | |
|
133 | 0 | const double dfPowerDiv2 = poOptions->dfPower / 2; |
134 | 0 | const double dfSmoothing = poOptions->dfSmoothing; |
135 | 0 | const GUInt32 nMaxPoints = poOptions->nMaxPoints; |
136 | 0 | double dfNominator = 0.0; |
137 | 0 | double dfDenominator = 0.0; |
138 | 0 | GUInt32 n = 0; |
139 | |
|
140 | 0 | for (GUInt32 i = 0; i < nPoints; i++) |
141 | 0 | { |
142 | 0 | double dfRX = padfX[i] - dfXPoint; |
143 | 0 | double dfRY = padfY[i] - dfYPoint; |
144 | 0 | const double dfR2 = |
145 | 0 | dfRX * dfRX + dfRY * dfRY + dfSmoothing * dfSmoothing; |
146 | |
|
147 | 0 | if (bRotated) |
148 | 0 | { |
149 | 0 | const double dfRXRotated = dfRX * dfCoeff1 + dfRY * dfCoeff2; |
150 | 0 | const double dfRYRotated = dfRY * dfCoeff1 - dfRX * dfCoeff2; |
151 | |
|
152 | 0 | dfRX = dfRXRotated; |
153 | 0 | dfRY = dfRYRotated; |
154 | 0 | } |
155 | | |
156 | | // Is this point located inside the search ellipse? |
157 | 0 | if (dfRadius2Square * dfRX * dfRX + dfRadius1Square * dfRY * dfRY <= |
158 | 0 | dfR12Square) |
159 | 0 | { |
160 | | // If the test point is close to the grid node, use the point |
161 | | // value directly as a node value to avoid singularity. |
162 | 0 | if (dfR2 < 0.0000000000001) |
163 | 0 | { |
164 | 0 | *pdfValue = padfZ[i]; |
165 | 0 | return CE_None; |
166 | 0 | } |
167 | | |
168 | 0 | const double dfW = pow(dfR2, dfPowerDiv2); |
169 | 0 | const double dfInvW = 1.0 / dfW; |
170 | 0 | dfNominator += dfInvW * padfZ[i]; |
171 | 0 | dfDenominator += dfInvW; |
172 | 0 | n++; |
173 | 0 | if (nMaxPoints > 0 && n > nMaxPoints) |
174 | 0 | break; |
175 | 0 | } |
176 | 0 | } |
177 | | |
178 | 0 | if (n < poOptions->nMinPoints || dfDenominator == 0.0) |
179 | 0 | { |
180 | 0 | *pdfValue = poOptions->dfNoDataValue; |
181 | 0 | } |
182 | 0 | else |
183 | 0 | { |
184 | 0 | *pdfValue = dfNominator / dfDenominator; |
185 | 0 | } |
186 | |
|
187 | 0 | return CE_None; |
188 | 0 | } |
189 | | |
190 | | /************************************************************************/ |
191 | | /* GDALGridInverseDistanceToAPowerNearestNeighbor() */ |
192 | | /************************************************************************/ |
193 | | |
194 | | /** |
195 | | * Inverse distance to a power with nearest neighbor search, ideal when |
196 | | * max_points used. |
197 | | * |
198 | | * The Inverse Distance to a Power gridding method is a weighted average |
199 | | * interpolator. You should supply the input arrays with the scattered data |
200 | | * values including coordinates of every data point and output grid geometry. |
201 | | * The function will compute interpolated value for the given position in |
202 | | * output grid. |
203 | | * |
204 | | * For every grid node the resulting value \f$Z\f$ will be calculated using |
205 | | * formula for nearest matches: |
206 | | * |
207 | | * \f[ |
208 | | * Z=\frac{\sum_{i=1}^n{\frac{Z_i}{r_i^p}}}{\sum_{i=1}^n{\frac{1}{r_i^p}}} |
209 | | * \f] |
210 | | * |
211 | | * where |
212 | | * <ul> |
213 | | * <li> \f$Z_i\f$ is a known value at point \f$i\f$, |
214 | | * <li> \f$r_i\f$ is an Euclidean distance from the grid node |
215 | | * to point \f$i\f$ (with an optional smoothing parameter \f$s\f$), |
216 | | * <li> \f$p\f$ is a weighting power, |
217 | | * <li> \f$n\f$ is a total number of points in search ellipse. |
218 | | * </ul> |
219 | | * |
220 | | * In this method the weighting factor \f$w\f$ is |
221 | | * |
222 | | * \f[ |
223 | | * w=\frac{1}{r^p} |
224 | | * \f] |
225 | | * |
226 | | * @param poOptionsIn Algorithm parameters. This should point to |
227 | | * GDALGridInverseDistanceToAPowerNearestNeighborOptions object. |
228 | | * @param nPoints Number of elements in input arrays. |
229 | | * @param padfX Input array of X coordinates. |
230 | | * @param padfY Input array of Y coordinates. |
231 | | * @param padfZ Input array of Z values. |
232 | | * @param dfXPoint X coordinate of the point to compute. |
233 | | * @param dfYPoint Y coordinate of the point to compute. |
234 | | * @param pdfValue Pointer to variable where the computed grid node value |
235 | | * will be returned. |
236 | | * @param hExtraParamsIn extra parameters. |
237 | | * |
238 | | * @return CE_None on success or CE_Failure if something goes wrong. |
239 | | */ |
240 | | |
241 | | CPLErr GDALGridInverseDistanceToAPowerNearestNeighbor( |
242 | | const void *poOptionsIn, GUInt32 nPoints, const double *padfX, |
243 | | const double *padfY, const double *padfZ, double dfXPoint, double dfYPoint, |
244 | | double *pdfValue, void *hExtraParamsIn) |
245 | 0 | { |
246 | 0 | CPL_IGNORE_RET_VAL(nPoints); |
247 | |
|
248 | 0 | const GDALGridInverseDistanceToAPowerNearestNeighborOptions |
249 | 0 | *const poOptions = static_cast< |
250 | 0 | const GDALGridInverseDistanceToAPowerNearestNeighborOptions *>( |
251 | 0 | poOptionsIn); |
252 | 0 | const double dfRadius = poOptions->dfRadius; |
253 | 0 | const double dfSmoothing = poOptions->dfSmoothing; |
254 | 0 | const double dfSmoothing2 = dfSmoothing * dfSmoothing; |
255 | |
|
256 | 0 | const GUInt32 nMaxPoints = poOptions->nMaxPoints; |
257 | |
|
258 | 0 | GDALGridExtraParameters *psExtraParams = |
259 | 0 | static_cast<GDALGridExtraParameters *>(hExtraParamsIn); |
260 | 0 | const CPLQuadTree *phQuadTree = psExtraParams->hQuadTree; |
261 | 0 | CPLAssert(phQuadTree); |
262 | |
|
263 | 0 | const double dfRPower2 = psExtraParams->dfRadiusPower2PreComp; |
264 | 0 | const double dfPowerDiv2 = psExtraParams->dfPowerDiv2PreComp; |
265 | |
|
266 | 0 | std::multimap<double, double> oMapDistanceToZValues; |
267 | |
|
268 | 0 | const double dfSearchRadius = dfRadius; |
269 | 0 | CPLRectObj sAoi; |
270 | 0 | sAoi.minx = dfXPoint - dfSearchRadius; |
271 | 0 | sAoi.miny = dfYPoint - dfSearchRadius; |
272 | 0 | sAoi.maxx = dfXPoint + dfSearchRadius; |
273 | 0 | sAoi.maxy = dfYPoint + dfSearchRadius; |
274 | 0 | int nFeatureCount = 0; |
275 | 0 | GDALGridPoint **papsPoints = reinterpret_cast<GDALGridPoint **>( |
276 | 0 | CPLQuadTreeSearch(phQuadTree, &sAoi, &nFeatureCount)); |
277 | 0 | if (nFeatureCount != 0) |
278 | 0 | { |
279 | 0 | for (int k = 0; k < nFeatureCount; k++) |
280 | 0 | { |
281 | 0 | const int i = papsPoints[k]->i; |
282 | 0 | const double dfRX = padfX[i] - dfXPoint; |
283 | 0 | const double dfRY = padfY[i] - dfYPoint; |
284 | |
|
285 | 0 | const double dfR2 = dfRX * dfRX + dfRY * dfRY; |
286 | | // real distance + smoothing |
287 | 0 | const double dfRsmoothed2 = dfR2 + dfSmoothing2; |
288 | 0 | if (dfRsmoothed2 < 0.0000000000001) |
289 | 0 | { |
290 | 0 | *pdfValue = padfZ[i]; |
291 | 0 | CPLFree(papsPoints); |
292 | 0 | return CE_None; |
293 | 0 | } |
294 | | // is point within real distance? |
295 | 0 | if (dfR2 <= dfRPower2) |
296 | 0 | { |
297 | 0 | oMapDistanceToZValues.insert( |
298 | 0 | std::make_pair(dfRsmoothed2, padfZ[i])); |
299 | 0 | } |
300 | 0 | } |
301 | 0 | } |
302 | 0 | CPLFree(papsPoints); |
303 | |
|
304 | 0 | double dfNominator = 0.0; |
305 | 0 | double dfDenominator = 0.0; |
306 | 0 | GUInt32 n = 0; |
307 | | |
308 | | // Examine all "neighbors" within the radius (sorted by distance via the |
309 | | // multimap), and use the closest n points based on distance until the max |
310 | | // is reached. |
311 | 0 | for (std::multimap<double, double>::iterator oMapDistanceToZValuesIter = |
312 | 0 | oMapDistanceToZValues.begin(); |
313 | 0 | oMapDistanceToZValuesIter != oMapDistanceToZValues.end(); |
314 | 0 | ++oMapDistanceToZValuesIter) |
315 | 0 | { |
316 | 0 | const double dfR2 = oMapDistanceToZValuesIter->first; |
317 | 0 | const double dfZ = oMapDistanceToZValuesIter->second; |
318 | |
|
319 | 0 | const double dfW = pow(dfR2, dfPowerDiv2); |
320 | 0 | const double dfInvW = 1.0 / dfW; |
321 | 0 | dfNominator += dfInvW * dfZ; |
322 | 0 | dfDenominator += dfInvW; |
323 | 0 | n++; |
324 | 0 | if (nMaxPoints > 0 && n >= nMaxPoints) |
325 | 0 | { |
326 | 0 | break; |
327 | 0 | } |
328 | 0 | } |
329 | |
|
330 | 0 | if (n < poOptions->nMinPoints || dfDenominator == 0.0) |
331 | 0 | { |
332 | 0 | *pdfValue = poOptions->dfNoDataValue; |
333 | 0 | } |
334 | 0 | else |
335 | 0 | { |
336 | 0 | *pdfValue = dfNominator / dfDenominator; |
337 | 0 | } |
338 | |
|
339 | 0 | return CE_None; |
340 | 0 | } |
341 | | |
342 | | /************************************************************************/ |
343 | | /* GDALGridInverseDistanceToAPowerNearestNeighborPerQuadrant() */ |
344 | | /************************************************************************/ |
345 | | |
346 | | /** |
347 | | * Inverse distance to a power with nearest neighbor search, with a per-quadrant |
348 | | * search logic. |
349 | | */ |
350 | | static CPLErr GDALGridInverseDistanceToAPowerNearestNeighborPerQuadrant( |
351 | | const void *poOptionsIn, GUInt32 /*nPoints*/, const double *padfX, |
352 | | const double *padfY, const double *padfZ, double dfXPoint, double dfYPoint, |
353 | | double *pdfValue, void *hExtraParamsIn) |
354 | 0 | { |
355 | 0 | const GDALGridInverseDistanceToAPowerNearestNeighborOptions |
356 | 0 | *const poOptions = static_cast< |
357 | 0 | const GDALGridInverseDistanceToAPowerNearestNeighborOptions *>( |
358 | 0 | poOptionsIn); |
359 | 0 | const double dfRadius = poOptions->dfRadius; |
360 | 0 | const double dfSmoothing = poOptions->dfSmoothing; |
361 | 0 | const double dfSmoothing2 = dfSmoothing * dfSmoothing; |
362 | |
|
363 | 0 | const GUInt32 nMaxPoints = poOptions->nMaxPoints; |
364 | 0 | const GUInt32 nMinPointsPerQuadrant = poOptions->nMinPointsPerQuadrant; |
365 | 0 | const GUInt32 nMaxPointsPerQuadrant = poOptions->nMaxPointsPerQuadrant; |
366 | |
|
367 | 0 | GDALGridExtraParameters *psExtraParams = |
368 | 0 | static_cast<GDALGridExtraParameters *>(hExtraParamsIn); |
369 | 0 | const CPLQuadTree *phQuadTree = psExtraParams->hQuadTree; |
370 | 0 | CPLAssert(phQuadTree); |
371 | |
|
372 | 0 | const double dfRPower2 = psExtraParams->dfRadiusPower2PreComp; |
373 | 0 | const double dfPowerDiv2 = psExtraParams->dfPowerDiv2PreComp; |
374 | 0 | std::multimap<double, double> oMapDistanceToZValuesPerQuadrant[4]; |
375 | |
|
376 | 0 | const double dfSearchRadius = dfRadius; |
377 | 0 | CPLRectObj sAoi; |
378 | 0 | sAoi.minx = dfXPoint - dfSearchRadius; |
379 | 0 | sAoi.miny = dfYPoint - dfSearchRadius; |
380 | 0 | sAoi.maxx = dfXPoint + dfSearchRadius; |
381 | 0 | sAoi.maxy = dfYPoint + dfSearchRadius; |
382 | 0 | int nFeatureCount = 0; |
383 | 0 | GDALGridPoint **papsPoints = reinterpret_cast<GDALGridPoint **>( |
384 | 0 | CPLQuadTreeSearch(phQuadTree, &sAoi, &nFeatureCount)); |
385 | 0 | if (nFeatureCount != 0) |
386 | 0 | { |
387 | 0 | for (int k = 0; k < nFeatureCount; k++) |
388 | 0 | { |
389 | 0 | const int i = papsPoints[k]->i; |
390 | 0 | const double dfRX = padfX[i] - dfXPoint; |
391 | 0 | const double dfRY = padfY[i] - dfYPoint; |
392 | |
|
393 | 0 | const double dfR2 = dfRX * dfRX + dfRY * dfRY; |
394 | | // real distance + smoothing |
395 | 0 | const double dfRsmoothed2 = dfR2 + dfSmoothing2; |
396 | 0 | if (dfRsmoothed2 < 0.0000000000001) |
397 | 0 | { |
398 | 0 | *pdfValue = padfZ[i]; |
399 | 0 | CPLFree(papsPoints); |
400 | 0 | return CE_None; |
401 | 0 | } |
402 | | // is point within real distance? |
403 | 0 | if (dfR2 <= dfRPower2) |
404 | 0 | { |
405 | 0 | const int iQuadrant = |
406 | 0 | ((dfRX >= 0) ? 1 : 0) | (((dfRY >= 0) ? 1 : 0) << 1); |
407 | 0 | oMapDistanceToZValuesPerQuadrant[iQuadrant].insert( |
408 | 0 | std::make_pair(dfRsmoothed2, padfZ[i])); |
409 | 0 | } |
410 | 0 | } |
411 | 0 | } |
412 | 0 | CPLFree(papsPoints); |
413 | |
|
414 | 0 | std::multimap<double, double>::iterator aoIter[] = { |
415 | 0 | oMapDistanceToZValuesPerQuadrant[0].begin(), |
416 | 0 | oMapDistanceToZValuesPerQuadrant[1].begin(), |
417 | 0 | oMapDistanceToZValuesPerQuadrant[2].begin(), |
418 | 0 | oMapDistanceToZValuesPerQuadrant[3].begin(), |
419 | 0 | }; |
420 | 0 | constexpr int ALL_QUADRANT_FLAGS = 1 + 2 + 4 + 8; |
421 | | |
422 | | // Examine all "neighbors" within the radius (sorted by distance via the |
423 | | // multimap), and use the closest n points based on distance until the max |
424 | | // is reached. |
425 | | // Do that by fetching the nearest point in quadrant 0, then the nearest |
426 | | // point in quadrant 1, 2 and 3, and starting again with the next nearest |
427 | | // point in quarant 0, etc. |
428 | 0 | int nQuadrantIterFinishedFlag = 0; |
429 | 0 | GUInt32 anPerQuadrant[4] = {0}; |
430 | 0 | double dfNominator = 0.0; |
431 | 0 | double dfDenominator = 0.0; |
432 | 0 | GUInt32 n = 0; |
433 | 0 | for (int iQuadrant = 0; /* true */; iQuadrant = (iQuadrant + 1) % 4) |
434 | 0 | { |
435 | 0 | if (aoIter[iQuadrant] == |
436 | 0 | oMapDistanceToZValuesPerQuadrant[iQuadrant].end() || |
437 | 0 | (nMaxPointsPerQuadrant > 0 && |
438 | 0 | anPerQuadrant[iQuadrant] >= nMaxPointsPerQuadrant)) |
439 | 0 | { |
440 | 0 | nQuadrantIterFinishedFlag |= 1 << iQuadrant; |
441 | 0 | if (nQuadrantIterFinishedFlag == ALL_QUADRANT_FLAGS) |
442 | 0 | break; |
443 | 0 | continue; |
444 | 0 | } |
445 | | |
446 | 0 | const double dfR2 = aoIter[iQuadrant]->first; |
447 | 0 | const double dfZ = aoIter[iQuadrant]->second; |
448 | 0 | ++aoIter[iQuadrant]; |
449 | |
|
450 | 0 | const double dfW = pow(dfR2, dfPowerDiv2); |
451 | 0 | const double dfInvW = 1.0 / dfW; |
452 | 0 | dfNominator += dfInvW * dfZ; |
453 | 0 | dfDenominator += dfInvW; |
454 | 0 | n++; |
455 | 0 | anPerQuadrant[iQuadrant]++; |
456 | 0 | if (nMaxPoints > 0 && n >= nMaxPoints) |
457 | 0 | { |
458 | 0 | break; |
459 | 0 | } |
460 | 0 | } |
461 | |
|
462 | 0 | if (nMinPointsPerQuadrant > 0 && |
463 | 0 | (anPerQuadrant[0] < nMinPointsPerQuadrant || |
464 | 0 | anPerQuadrant[1] < nMinPointsPerQuadrant || |
465 | 0 | anPerQuadrant[2] < nMinPointsPerQuadrant || |
466 | 0 | anPerQuadrant[3] < nMinPointsPerQuadrant)) |
467 | 0 | { |
468 | 0 | *pdfValue = poOptions->dfNoDataValue; |
469 | 0 | } |
470 | 0 | else if (n < poOptions->nMinPoints || dfDenominator == 0.0) |
471 | 0 | { |
472 | 0 | *pdfValue = poOptions->dfNoDataValue; |
473 | 0 | } |
474 | 0 | else |
475 | 0 | { |
476 | 0 | *pdfValue = dfNominator / dfDenominator; |
477 | 0 | } |
478 | |
|
479 | 0 | return CE_None; |
480 | 0 | } |
481 | | |
482 | | /************************************************************************/ |
483 | | /* GDALGridInverseDistanceToAPowerNoSearch() */ |
484 | | /************************************************************************/ |
485 | | |
486 | | /** |
487 | | * Inverse distance to a power for whole data set. |
488 | | * |
489 | | * This is somewhat optimized version of the Inverse Distance to a Power |
490 | | * method. It is used when the search ellips is not set. The algorithm and |
491 | | * parameters are the same as in GDALGridInverseDistanceToAPower(), but this |
492 | | * implementation works faster, because of no search. |
493 | | * |
494 | | * @see GDALGridInverseDistanceToAPower() |
495 | | */ |
496 | | |
497 | | CPLErr GDALGridInverseDistanceToAPowerNoSearch( |
498 | | const void *poOptionsIn, GUInt32 nPoints, const double *padfX, |
499 | | const double *padfY, const double *padfZ, double dfXPoint, double dfYPoint, |
500 | | double *pdfValue, void * /* hExtraParamsIn */) |
501 | 0 | { |
502 | 0 | const GDALGridInverseDistanceToAPowerOptions *const poOptions = |
503 | 0 | static_cast<const GDALGridInverseDistanceToAPowerOptions *>( |
504 | 0 | poOptionsIn); |
505 | 0 | const double dfPowerDiv2 = poOptions->dfPower / 2.0; |
506 | 0 | const double dfSmoothing = poOptions->dfSmoothing; |
507 | 0 | const double dfSmoothing2 = dfSmoothing * dfSmoothing; |
508 | 0 | double dfNominator = 0.0; |
509 | 0 | double dfDenominator = 0.0; |
510 | 0 | const bool bPower2 = dfPowerDiv2 == 1.0; |
511 | |
|
512 | 0 | GUInt32 i = 0; // Used after if. |
513 | 0 | if (bPower2) |
514 | 0 | { |
515 | 0 | if (dfSmoothing2 > 0) |
516 | 0 | { |
517 | 0 | for (i = 0; i < nPoints; i++) |
518 | 0 | { |
519 | 0 | const double dfRX = padfX[i] - dfXPoint; |
520 | 0 | const double dfRY = padfY[i] - dfYPoint; |
521 | 0 | const double dfR2 = dfRX * dfRX + dfRY * dfRY + dfSmoothing2; |
522 | |
|
523 | 0 | const double dfInvR2 = 1.0 / dfR2; |
524 | 0 | dfNominator += dfInvR2 * padfZ[i]; |
525 | 0 | dfDenominator += dfInvR2; |
526 | 0 | } |
527 | 0 | } |
528 | 0 | else |
529 | 0 | { |
530 | 0 | for (i = 0; i < nPoints; i++) |
531 | 0 | { |
532 | 0 | const double dfRX = padfX[i] - dfXPoint; |
533 | 0 | const double dfRY = padfY[i] - dfYPoint; |
534 | 0 | const double dfR2 = dfRX * dfRX + dfRY * dfRY; |
535 | | |
536 | | // If the test point is close to the grid node, use the point |
537 | | // value directly as a node value to avoid singularity. |
538 | 0 | if (dfR2 < 0.0000000000001) |
539 | 0 | { |
540 | 0 | break; |
541 | 0 | } |
542 | | |
543 | 0 | const double dfInvR2 = 1.0 / dfR2; |
544 | 0 | dfNominator += dfInvR2 * padfZ[i]; |
545 | 0 | dfDenominator += dfInvR2; |
546 | 0 | } |
547 | 0 | } |
548 | 0 | } |
549 | 0 | else |
550 | 0 | { |
551 | 0 | for (i = 0; i < nPoints; i++) |
552 | 0 | { |
553 | 0 | const double dfRX = padfX[i] - dfXPoint; |
554 | 0 | const double dfRY = padfY[i] - dfYPoint; |
555 | 0 | const double dfR2 = dfRX * dfRX + dfRY * dfRY + dfSmoothing2; |
556 | | |
557 | | // If the test point is close to the grid node, use the point |
558 | | // value directly as a node value to avoid singularity. |
559 | 0 | if (dfR2 < 0.0000000000001) |
560 | 0 | { |
561 | 0 | break; |
562 | 0 | } |
563 | | |
564 | 0 | const double dfW = pow(dfR2, dfPowerDiv2); |
565 | 0 | const double dfInvW = 1.0 / dfW; |
566 | 0 | dfNominator += dfInvW * padfZ[i]; |
567 | 0 | dfDenominator += dfInvW; |
568 | 0 | } |
569 | 0 | } |
570 | |
|
571 | 0 | if (i != nPoints) |
572 | 0 | { |
573 | 0 | *pdfValue = padfZ[i]; |
574 | 0 | } |
575 | 0 | else if (dfDenominator == 0.0) |
576 | 0 | { |
577 | 0 | *pdfValue = poOptions->dfNoDataValue; |
578 | 0 | } |
579 | 0 | else |
580 | 0 | { |
581 | 0 | *pdfValue = dfNominator / dfDenominator; |
582 | 0 | } |
583 | |
|
584 | 0 | return CE_None; |
585 | 0 | } |
586 | | |
587 | | /************************************************************************/ |
588 | | /* GDALGridMovingAverage() */ |
589 | | /************************************************************************/ |
590 | | |
591 | | /** |
592 | | * Moving average. |
593 | | * |
594 | | * The Moving Average is a simple data averaging algorithm. It uses a moving |
595 | | * window of elliptic form to search values and averages all data points |
596 | | * within the window. Search ellipse can be rotated by specified angle, the |
597 | | * center of ellipse located at the grid node. Also the minimum number of data |
598 | | * points to average can be set, if there are not enough points in window, the |
599 | | * grid node considered empty and will be filled with specified NODATA value. |
600 | | * |
601 | | * Mathematically it can be expressed with the formula: |
602 | | * |
603 | | * \f[ |
604 | | * Z=\frac{\sum_{i=1}^n{Z_i}}{n} |
605 | | * \f] |
606 | | * |
607 | | * where |
608 | | * <ul> |
609 | | * <li> \f$Z\f$ is a resulting value at the grid node, |
610 | | * <li> \f$Z_i\f$ is a known value at point \f$i\f$, |
611 | | * <li> \f$n\f$ is a total number of points in search ellipse. |
612 | | * </ul> |
613 | | * |
614 | | * @param poOptionsIn Algorithm parameters. This should point to |
615 | | * GDALGridMovingAverageOptions object. |
616 | | * @param nPoints Number of elements in input arrays. |
617 | | * @param padfX Input array of X coordinates. |
618 | | * @param padfY Input array of Y coordinates. |
619 | | * @param padfZ Input array of Z values. |
620 | | * @param dfXPoint X coordinate of the point to compute. |
621 | | * @param dfYPoint Y coordinate of the point to compute. |
622 | | * @param pdfValue Pointer to variable where the computed grid node value |
623 | | * will be returned. |
624 | | * @param hExtraParamsIn extra parameters (unused) |
625 | | * |
626 | | * @return CE_None on success or CE_Failure if something goes wrong. |
627 | | */ |
628 | | |
629 | | CPLErr GDALGridMovingAverage(const void *poOptionsIn, GUInt32 nPoints, |
630 | | const double *padfX, const double *padfY, |
631 | | const double *padfZ, double dfXPoint, |
632 | | double dfYPoint, double *pdfValue, |
633 | | CPL_UNUSED void *hExtraParamsIn) |
634 | 0 | { |
635 | | // TODO: For optimization purposes pre-computed parameters should be moved |
636 | | // out of this routine to the calling function. |
637 | |
|
638 | 0 | const GDALGridMovingAverageOptions *const poOptions = |
639 | 0 | static_cast<const GDALGridMovingAverageOptions *>(poOptionsIn); |
640 | | // Pre-compute search ellipse parameters. |
641 | 0 | const double dfRadius1Square = poOptions->dfRadius1 * poOptions->dfRadius1; |
642 | 0 | const double dfRadius2Square = poOptions->dfRadius2 * poOptions->dfRadius2; |
643 | 0 | const double dfSearchRadius = |
644 | 0 | std::max(poOptions->dfRadius1, poOptions->dfRadius2); |
645 | 0 | const double dfR12Square = dfRadius1Square * dfRadius2Square; |
646 | |
|
647 | 0 | GDALGridExtraParameters *psExtraParams = |
648 | 0 | static_cast<GDALGridExtraParameters *>(hExtraParamsIn); |
649 | 0 | const CPLQuadTree *phQuadTree = psExtraParams->hQuadTree; |
650 | | |
651 | | // Compute coefficients for coordinate system rotation. |
652 | 0 | const double dfAngle = TO_RADIANS * poOptions->dfAngle; |
653 | 0 | const bool bRotated = dfAngle != 0.0; |
654 | |
|
655 | 0 | const double dfCoeff1 = bRotated ? cos(dfAngle) : 0.0; |
656 | 0 | const double dfCoeff2 = bRotated ? sin(dfAngle) : 0.0; |
657 | |
|
658 | 0 | double dfAccumulator = 0.0; |
659 | |
|
660 | 0 | GUInt32 n = 0; // Used after for. |
661 | 0 | if (phQuadTree != nullptr) |
662 | 0 | { |
663 | 0 | CPLRectObj sAoi; |
664 | 0 | sAoi.minx = dfXPoint - dfSearchRadius; |
665 | 0 | sAoi.miny = dfYPoint - dfSearchRadius; |
666 | 0 | sAoi.maxx = dfXPoint + dfSearchRadius; |
667 | 0 | sAoi.maxy = dfYPoint + dfSearchRadius; |
668 | 0 | int nFeatureCount = 0; |
669 | 0 | GDALGridPoint **papsPoints = reinterpret_cast<GDALGridPoint **>( |
670 | 0 | CPLQuadTreeSearch(phQuadTree, &sAoi, &nFeatureCount)); |
671 | 0 | if (nFeatureCount != 0) |
672 | 0 | { |
673 | 0 | for (int k = 0; k < nFeatureCount; k++) |
674 | 0 | { |
675 | 0 | const int i = papsPoints[k]->i; |
676 | 0 | const double dfRX = padfX[i] - dfXPoint; |
677 | 0 | const double dfRY = padfY[i] - dfYPoint; |
678 | |
|
679 | 0 | if (dfRadius2Square * dfRX * dfRX + |
680 | 0 | dfRadius1Square * dfRY * dfRY <= |
681 | 0 | dfR12Square) |
682 | 0 | { |
683 | 0 | dfAccumulator += padfZ[i]; |
684 | 0 | n++; |
685 | 0 | } |
686 | 0 | } |
687 | 0 | } |
688 | 0 | CPLFree(papsPoints); |
689 | 0 | } |
690 | 0 | else |
691 | 0 | { |
692 | 0 | for (GUInt32 i = 0; i < nPoints; i++) |
693 | 0 | { |
694 | 0 | double dfRX = padfX[i] - dfXPoint; |
695 | 0 | double dfRY = padfY[i] - dfYPoint; |
696 | |
|
697 | 0 | if (bRotated) |
698 | 0 | { |
699 | 0 | const double dfRXRotated = dfRX * dfCoeff1 + dfRY * dfCoeff2; |
700 | 0 | const double dfRYRotated = dfRY * dfCoeff1 - dfRX * dfCoeff2; |
701 | |
|
702 | 0 | dfRX = dfRXRotated; |
703 | 0 | dfRY = dfRYRotated; |
704 | 0 | } |
705 | | |
706 | | // Is this point located inside the search ellipse? |
707 | 0 | if (dfRadius2Square * dfRX * dfRX + dfRadius1Square * dfRY * dfRY <= |
708 | 0 | dfR12Square) |
709 | 0 | { |
710 | 0 | dfAccumulator += padfZ[i]; |
711 | 0 | n++; |
712 | 0 | } |
713 | 0 | } |
714 | 0 | } |
715 | |
|
716 | 0 | if (n < poOptions->nMinPoints || n == 0) |
717 | 0 | { |
718 | 0 | *pdfValue = poOptions->dfNoDataValue; |
719 | 0 | } |
720 | 0 | else |
721 | 0 | { |
722 | 0 | *pdfValue = dfAccumulator / n; |
723 | 0 | } |
724 | |
|
725 | 0 | return CE_None; |
726 | 0 | } |
727 | | |
728 | | /************************************************************************/ |
729 | | /* GDALGridMovingAveragePerQuadrant() */ |
730 | | /************************************************************************/ |
731 | | |
732 | | /** |
733 | | * Moving average, with a per-quadrant search logic. |
734 | | */ |
735 | | static CPLErr GDALGridMovingAveragePerQuadrant( |
736 | | const void *poOptionsIn, GUInt32 /*nPoints*/, const double *padfX, |
737 | | const double *padfY, const double *padfZ, double dfXPoint, double dfYPoint, |
738 | | double *pdfValue, void *hExtraParamsIn) |
739 | 0 | { |
740 | 0 | const GDALGridMovingAverageOptions *const poOptions = |
741 | 0 | static_cast<const GDALGridMovingAverageOptions *>(poOptionsIn); |
742 | 0 | const double dfRadius1Square = poOptions->dfRadius1 * poOptions->dfRadius1; |
743 | 0 | const double dfRadius2Square = poOptions->dfRadius2 * poOptions->dfRadius2; |
744 | 0 | const double dfR12Square = dfRadius1Square * dfRadius2Square; |
745 | |
|
746 | 0 | const GUInt32 nMaxPoints = poOptions->nMaxPoints; |
747 | 0 | const GUInt32 nMinPointsPerQuadrant = poOptions->nMinPointsPerQuadrant; |
748 | 0 | const GUInt32 nMaxPointsPerQuadrant = poOptions->nMaxPointsPerQuadrant; |
749 | |
|
750 | 0 | GDALGridExtraParameters *psExtraParams = |
751 | 0 | static_cast<GDALGridExtraParameters *>(hExtraParamsIn); |
752 | 0 | const CPLQuadTree *phQuadTree = psExtraParams->hQuadTree; |
753 | 0 | CPLAssert(phQuadTree); |
754 | |
|
755 | 0 | std::multimap<double, double> oMapDistanceToZValuesPerQuadrant[4]; |
756 | |
|
757 | 0 | const double dfSearchRadius = |
758 | 0 | std::max(poOptions->dfRadius1, poOptions->dfRadius2); |
759 | 0 | CPLRectObj sAoi; |
760 | 0 | sAoi.minx = dfXPoint - dfSearchRadius; |
761 | 0 | sAoi.miny = dfYPoint - dfSearchRadius; |
762 | 0 | sAoi.maxx = dfXPoint + dfSearchRadius; |
763 | 0 | sAoi.maxy = dfYPoint + dfSearchRadius; |
764 | 0 | int nFeatureCount = 0; |
765 | 0 | GDALGridPoint **papsPoints = reinterpret_cast<GDALGridPoint **>( |
766 | 0 | CPLQuadTreeSearch(phQuadTree, &sAoi, &nFeatureCount)); |
767 | 0 | if (nFeatureCount != 0) |
768 | 0 | { |
769 | 0 | for (int k = 0; k < nFeatureCount; k++) |
770 | 0 | { |
771 | 0 | const int i = papsPoints[k]->i; |
772 | 0 | const double dfRX = padfX[i] - dfXPoint; |
773 | 0 | const double dfRY = padfY[i] - dfYPoint; |
774 | 0 | const double dfRXSquare = dfRX * dfRX; |
775 | 0 | const double dfRYSquare = dfRY * dfRY; |
776 | |
|
777 | 0 | if (dfRadius2Square * dfRXSquare + dfRadius1Square * dfRYSquare <= |
778 | 0 | dfR12Square) |
779 | 0 | { |
780 | 0 | const int iQuadrant = |
781 | 0 | ((dfRX >= 0) ? 1 : 0) | (((dfRY >= 0) ? 1 : 0) << 1); |
782 | 0 | oMapDistanceToZValuesPerQuadrant[iQuadrant].insert( |
783 | 0 | std::make_pair(dfRXSquare + dfRYSquare, padfZ[i])); |
784 | 0 | } |
785 | 0 | } |
786 | 0 | } |
787 | 0 | CPLFree(papsPoints); |
788 | |
|
789 | 0 | std::multimap<double, double>::iterator aoIter[] = { |
790 | 0 | oMapDistanceToZValuesPerQuadrant[0].begin(), |
791 | 0 | oMapDistanceToZValuesPerQuadrant[1].begin(), |
792 | 0 | oMapDistanceToZValuesPerQuadrant[2].begin(), |
793 | 0 | oMapDistanceToZValuesPerQuadrant[3].begin(), |
794 | 0 | }; |
795 | 0 | constexpr int ALL_QUADRANT_FLAGS = 1 + 2 + 4 + 8; |
796 | | |
797 | | // Examine all "neighbors" within the radius (sorted by distance via the |
798 | | // multimap), and use the closest n points based on distance until the max |
799 | | // is reached. |
800 | | // Do that by fetching the nearest point in quadrant 0, then the nearest |
801 | | // point in quadrant 1, 2 and 3, and starting again with the next nearest |
802 | | // point in quarant 0, etc. |
803 | 0 | int nQuadrantIterFinishedFlag = 0; |
804 | 0 | GUInt32 anPerQuadrant[4] = {0}; |
805 | 0 | double dfNominator = 0.0; |
806 | 0 | GUInt32 n = 0; |
807 | 0 | for (int iQuadrant = 0; /* true */; iQuadrant = (iQuadrant + 1) % 4) |
808 | 0 | { |
809 | 0 | if (aoIter[iQuadrant] == |
810 | 0 | oMapDistanceToZValuesPerQuadrant[iQuadrant].end() || |
811 | 0 | (nMaxPointsPerQuadrant > 0 && |
812 | 0 | anPerQuadrant[iQuadrant] >= nMaxPointsPerQuadrant)) |
813 | 0 | { |
814 | 0 | nQuadrantIterFinishedFlag |= 1 << iQuadrant; |
815 | 0 | if (nQuadrantIterFinishedFlag == ALL_QUADRANT_FLAGS) |
816 | 0 | break; |
817 | 0 | continue; |
818 | 0 | } |
819 | | |
820 | 0 | const double dfZ = aoIter[iQuadrant]->second; |
821 | 0 | ++aoIter[iQuadrant]; |
822 | |
|
823 | 0 | dfNominator += dfZ; |
824 | 0 | n++; |
825 | 0 | anPerQuadrant[iQuadrant]++; |
826 | 0 | if (nMaxPoints > 0 && n >= nMaxPoints) |
827 | 0 | { |
828 | 0 | break; |
829 | 0 | } |
830 | 0 | } |
831 | |
|
832 | 0 | if (nMinPointsPerQuadrant > 0 && |
833 | 0 | (anPerQuadrant[0] < nMinPointsPerQuadrant || |
834 | 0 | anPerQuadrant[1] < nMinPointsPerQuadrant || |
835 | 0 | anPerQuadrant[2] < nMinPointsPerQuadrant || |
836 | 0 | anPerQuadrant[3] < nMinPointsPerQuadrant)) |
837 | 0 | { |
838 | 0 | *pdfValue = poOptions->dfNoDataValue; |
839 | 0 | } |
840 | 0 | else if (n < poOptions->nMinPoints || n == 0) |
841 | 0 | { |
842 | 0 | *pdfValue = poOptions->dfNoDataValue; |
843 | 0 | } |
844 | 0 | else |
845 | 0 | { |
846 | 0 | *pdfValue = dfNominator / n; |
847 | 0 | } |
848 | |
|
849 | 0 | return CE_None; |
850 | 0 | } |
851 | | |
852 | | /************************************************************************/ |
853 | | /* GDALGridNearestNeighbor() */ |
854 | | /************************************************************************/ |
855 | | |
856 | | /** |
857 | | * Nearest neighbor. |
858 | | * |
859 | | * The Nearest Neighbor method doesn't perform any interpolation or smoothing, |
860 | | * it just takes the value of nearest point found in grid node search ellipse |
861 | | * and returns it as a result. If there are no points found, the specified |
862 | | * NODATA value will be returned. |
863 | | * |
864 | | * @param poOptionsIn Algorithm parameters. This should point to |
865 | | * GDALGridNearestNeighborOptions object. |
866 | | * @param nPoints Number of elements in input arrays. |
867 | | * @param padfX Input array of X coordinates. |
868 | | * @param padfY Input array of Y coordinates. |
869 | | * @param padfZ Input array of Z values. |
870 | | * @param dfXPoint X coordinate of the point to compute. |
871 | | * @param dfYPoint Y coordinate of the point to compute. |
872 | | * @param pdfValue Pointer to variable where the computed grid node value |
873 | | * will be returned. |
874 | | * @param hExtraParamsIn extra parameters. |
875 | | * |
876 | | * @return CE_None on success or CE_Failure if something goes wrong. |
877 | | */ |
878 | | |
879 | | CPLErr GDALGridNearestNeighbor(const void *poOptionsIn, GUInt32 nPoints, |
880 | | const double *padfX, const double *padfY, |
881 | | const double *padfZ, double dfXPoint, |
882 | | double dfYPoint, double *pdfValue, |
883 | | void *hExtraParamsIn) |
884 | 0 | { |
885 | | // TODO: For optimization purposes pre-computed parameters should be moved |
886 | | // out of this routine to the calling function. |
887 | |
|
888 | 0 | const GDALGridNearestNeighborOptions *const poOptions = |
889 | 0 | static_cast<const GDALGridNearestNeighborOptions *>(poOptionsIn); |
890 | | // Pre-compute search ellipse parameters. |
891 | 0 | const double dfRadius1Square = poOptions->dfRadius1 * poOptions->dfRadius1; |
892 | 0 | const double dfRadius2Square = poOptions->dfRadius2 * poOptions->dfRadius2; |
893 | 0 | const double dfR12Square = dfRadius1Square * dfRadius2Square; |
894 | 0 | GDALGridExtraParameters *psExtraParams = |
895 | 0 | static_cast<GDALGridExtraParameters *>(hExtraParamsIn); |
896 | 0 | CPLQuadTree *hQuadTree = psExtraParams->hQuadTree; |
897 | | |
898 | | // Compute coefficients for coordinate system rotation. |
899 | 0 | const double dfAngle = TO_RADIANS * poOptions->dfAngle; |
900 | 0 | const bool bRotated = dfAngle != 0.0; |
901 | 0 | const double dfCoeff1 = bRotated ? cos(dfAngle) : 0.0; |
902 | 0 | const double dfCoeff2 = bRotated ? sin(dfAngle) : 0.0; |
903 | | |
904 | | // If the nearest point will not be found, its value remains as NODATA. |
905 | 0 | double dfNearestValue = poOptions->dfNoDataValue; |
906 | 0 | GUInt32 i = 0; |
907 | |
|
908 | 0 | double dfSearchRadius = psExtraParams->dfInitialSearchRadius; |
909 | 0 | if (hQuadTree != nullptr) |
910 | 0 | { |
911 | 0 | if (poOptions->dfRadius1 > 0 || poOptions->dfRadius2 > 0) |
912 | 0 | dfSearchRadius = |
913 | 0 | std::max(poOptions->dfRadius1, poOptions->dfRadius2); |
914 | 0 | CPLRectObj sAoi; |
915 | 0 | while (dfSearchRadius > 0) |
916 | 0 | { |
917 | 0 | sAoi.minx = dfXPoint - dfSearchRadius; |
918 | 0 | sAoi.miny = dfYPoint - dfSearchRadius; |
919 | 0 | sAoi.maxx = dfXPoint + dfSearchRadius; |
920 | 0 | sAoi.maxy = dfYPoint + dfSearchRadius; |
921 | 0 | int nFeatureCount = 0; |
922 | 0 | GDALGridPoint **papsPoints = reinterpret_cast<GDALGridPoint **>( |
923 | 0 | CPLQuadTreeSearch(hQuadTree, &sAoi, &nFeatureCount)); |
924 | 0 | if (nFeatureCount != 0) |
925 | 0 | { |
926 | | // Nearest distance will be initialized with the distance to the |
927 | | // first point in array. |
928 | 0 | double dfNearestRSquare = std::numeric_limits<double>::max(); |
929 | 0 | for (int k = 0; k < nFeatureCount; k++) |
930 | 0 | { |
931 | 0 | const int idx = papsPoints[k]->i; |
932 | 0 | const double dfRX = padfX[idx] - dfXPoint; |
933 | 0 | const double dfRY = padfY[idx] - dfYPoint; |
934 | |
|
935 | 0 | const double dfR2 = dfRX * dfRX + dfRY * dfRY; |
936 | 0 | if (dfR2 <= dfNearestRSquare) |
937 | 0 | { |
938 | 0 | dfNearestRSquare = dfR2; |
939 | 0 | dfNearestValue = padfZ[idx]; |
940 | 0 | } |
941 | 0 | } |
942 | |
|
943 | 0 | CPLFree(papsPoints); |
944 | 0 | break; |
945 | 0 | } |
946 | | |
947 | 0 | CPLFree(papsPoints); |
948 | 0 | if (poOptions->dfRadius1 > 0 || poOptions->dfRadius2 > 0) |
949 | 0 | break; |
950 | 0 | dfSearchRadius *= 2; |
951 | | #if DEBUG_VERBOSE |
952 | | CPLDebug("GDAL_GRID", "Increasing search radius to %.16g", |
953 | | dfSearchRadius); |
954 | | #endif |
955 | 0 | } |
956 | 0 | } |
957 | 0 | else |
958 | 0 | { |
959 | 0 | double dfNearestRSquare = std::numeric_limits<double>::max(); |
960 | 0 | while (i < nPoints) |
961 | 0 | { |
962 | 0 | double dfRX = padfX[i] - dfXPoint; |
963 | 0 | double dfRY = padfY[i] - dfYPoint; |
964 | |
|
965 | 0 | if (bRotated) |
966 | 0 | { |
967 | 0 | const double dfRXRotated = dfRX * dfCoeff1 + dfRY * dfCoeff2; |
968 | 0 | const double dfRYRotated = dfRY * dfCoeff1 - dfRX * dfCoeff2; |
969 | |
|
970 | 0 | dfRX = dfRXRotated; |
971 | 0 | dfRY = dfRYRotated; |
972 | 0 | } |
973 | | |
974 | | // Is this point located inside the search ellipse? |
975 | 0 | const double dfRXSquare = dfRX * dfRX; |
976 | 0 | const double dfRYSquare = dfRY * dfRY; |
977 | 0 | if (dfRadius2Square * dfRXSquare + dfRadius1Square * dfRYSquare <= |
978 | 0 | dfR12Square) |
979 | 0 | { |
980 | 0 | const double dfR2 = dfRXSquare + dfRYSquare; |
981 | 0 | if (dfR2 <= dfNearestRSquare) |
982 | 0 | { |
983 | 0 | dfNearestRSquare = dfR2; |
984 | 0 | dfNearestValue = padfZ[i]; |
985 | 0 | } |
986 | 0 | } |
987 | |
|
988 | 0 | i++; |
989 | 0 | } |
990 | 0 | } |
991 | |
|
992 | 0 | *pdfValue = dfNearestValue; |
993 | |
|
994 | 0 | return CE_None; |
995 | 0 | } |
996 | | |
997 | | /************************************************************************/ |
998 | | /* GDALGridDataMetricMinimum() */ |
999 | | /************************************************************************/ |
1000 | | |
1001 | | /** |
1002 | | * Minimum data value (data metric). |
1003 | | * |
1004 | | * Minimum value found in grid node search ellipse. If there are no points |
1005 | | * found, the specified NODATA value will be returned. |
1006 | | * |
1007 | | * \f[ |
1008 | | * Z=\min{(Z_1,Z_2,\ldots,Z_n)} |
1009 | | * \f] |
1010 | | * |
1011 | | * where |
1012 | | * <ul> |
1013 | | * <li> \f$Z\f$ is a resulting value at the grid node, |
1014 | | * <li> \f$Z_i\f$ is a known value at point \f$i\f$, |
1015 | | * <li> \f$n\f$ is a total number of points in search ellipse. |
1016 | | * </ul> |
1017 | | * |
1018 | | * @param poOptionsIn Algorithm parameters. This should point to |
1019 | | * GDALGridDataMetricsOptions object. |
1020 | | * @param nPoints Number of elements in input arrays. |
1021 | | * @param padfX Input array of X coordinates. |
1022 | | * @param padfY Input array of Y coordinates. |
1023 | | * @param padfZ Input array of Z values. |
1024 | | * @param dfXPoint X coordinate of the point to compute. |
1025 | | * @param dfYPoint Y coordinate of the point to compute. |
1026 | | * @param pdfValue Pointer to variable where the computed grid node value |
1027 | | * will be returned. |
1028 | | * @param hExtraParamsIn unused. |
1029 | | * |
1030 | | * @return CE_None on success or CE_Failure if something goes wrong. |
1031 | | */ |
1032 | | |
1033 | | CPLErr GDALGridDataMetricMinimum(const void *poOptionsIn, GUInt32 nPoints, |
1034 | | const double *padfX, const double *padfY, |
1035 | | const double *padfZ, double dfXPoint, |
1036 | | double dfYPoint, double *pdfValue, |
1037 | | void *hExtraParamsIn) |
1038 | 0 | { |
1039 | | // TODO: For optimization purposes pre-computed parameters should be moved |
1040 | | // out of this routine to the calling function. |
1041 | |
|
1042 | 0 | const GDALGridDataMetricsOptions *const poOptions = |
1043 | 0 | static_cast<const GDALGridDataMetricsOptions *>(poOptionsIn); |
1044 | | |
1045 | | // Pre-compute search ellipse parameters. |
1046 | 0 | const double dfRadius1Square = poOptions->dfRadius1 * poOptions->dfRadius1; |
1047 | 0 | const double dfRadius2Square = poOptions->dfRadius2 * poOptions->dfRadius2; |
1048 | 0 | const double dfSearchRadius = |
1049 | 0 | std::max(poOptions->dfRadius1, poOptions->dfRadius2); |
1050 | 0 | const double dfR12Square = dfRadius1Square * dfRadius2Square; |
1051 | |
|
1052 | 0 | GDALGridExtraParameters *psExtraParams = |
1053 | 0 | static_cast<GDALGridExtraParameters *>(hExtraParamsIn); |
1054 | 0 | CPLQuadTree *phQuadTree = psExtraParams->hQuadTree; |
1055 | | |
1056 | | // Compute coefficients for coordinate system rotation. |
1057 | 0 | const double dfAngle = TO_RADIANS * poOptions->dfAngle; |
1058 | 0 | const bool bRotated = dfAngle != 0.0; |
1059 | 0 | const double dfCoeff1 = bRotated ? cos(dfAngle) : 0.0; |
1060 | 0 | const double dfCoeff2 = bRotated ? sin(dfAngle) : 0.0; |
1061 | |
|
1062 | 0 | double dfMinimumValue = std::numeric_limits<double>::max(); |
1063 | 0 | GUInt32 n = 0; |
1064 | 0 | if (phQuadTree != nullptr) |
1065 | 0 | { |
1066 | 0 | CPLRectObj sAoi; |
1067 | 0 | sAoi.minx = dfXPoint - dfSearchRadius; |
1068 | 0 | sAoi.miny = dfYPoint - dfSearchRadius; |
1069 | 0 | sAoi.maxx = dfXPoint + dfSearchRadius; |
1070 | 0 | sAoi.maxy = dfYPoint + dfSearchRadius; |
1071 | 0 | int nFeatureCount = 0; |
1072 | 0 | GDALGridPoint **papsPoints = reinterpret_cast<GDALGridPoint **>( |
1073 | 0 | CPLQuadTreeSearch(phQuadTree, &sAoi, &nFeatureCount)); |
1074 | 0 | if (nFeatureCount != 0) |
1075 | 0 | { |
1076 | 0 | for (int k = 0; k < nFeatureCount; k++) |
1077 | 0 | { |
1078 | 0 | const int i = papsPoints[k]->i; |
1079 | 0 | const double dfRX = padfX[i] - dfXPoint; |
1080 | 0 | const double dfRY = padfY[i] - dfYPoint; |
1081 | |
|
1082 | 0 | if (dfRadius2Square * dfRX * dfRX + |
1083 | 0 | dfRadius1Square * dfRY * dfRY <= |
1084 | 0 | dfR12Square) |
1085 | 0 | { |
1086 | 0 | if (dfMinimumValue > padfZ[i]) |
1087 | 0 | dfMinimumValue = padfZ[i]; |
1088 | 0 | n++; |
1089 | 0 | } |
1090 | 0 | } |
1091 | 0 | } |
1092 | 0 | CPLFree(papsPoints); |
1093 | 0 | } |
1094 | 0 | else |
1095 | 0 | { |
1096 | 0 | GUInt32 i = 0; |
1097 | 0 | while (i < nPoints) |
1098 | 0 | { |
1099 | 0 | double dfRX = padfX[i] - dfXPoint; |
1100 | 0 | double dfRY = padfY[i] - dfYPoint; |
1101 | |
|
1102 | 0 | if (bRotated) |
1103 | 0 | { |
1104 | 0 | const double dfRXRotated = dfRX * dfCoeff1 + dfRY * dfCoeff2; |
1105 | 0 | const double dfRYRotated = dfRY * dfCoeff1 - dfRX * dfCoeff2; |
1106 | |
|
1107 | 0 | dfRX = dfRXRotated; |
1108 | 0 | dfRY = dfRYRotated; |
1109 | 0 | } |
1110 | | |
1111 | | // Is this point located inside the search ellipse? |
1112 | 0 | if (dfRadius2Square * dfRX * dfRX + dfRadius1Square * dfRY * dfRY <= |
1113 | 0 | dfR12Square) |
1114 | 0 | { |
1115 | 0 | if (dfMinimumValue > padfZ[i]) |
1116 | 0 | dfMinimumValue = padfZ[i]; |
1117 | 0 | n++; |
1118 | 0 | } |
1119 | |
|
1120 | 0 | i++; |
1121 | 0 | } |
1122 | 0 | } |
1123 | |
|
1124 | 0 | if (n < poOptions->nMinPoints || n == 0) |
1125 | 0 | { |
1126 | 0 | *pdfValue = poOptions->dfNoDataValue; |
1127 | 0 | } |
1128 | 0 | else |
1129 | 0 | { |
1130 | 0 | *pdfValue = dfMinimumValue; |
1131 | 0 | } |
1132 | |
|
1133 | 0 | return CE_None; |
1134 | 0 | } |
1135 | | |
1136 | | /************************************************************************/ |
1137 | | /* GDALGridDataMetricMinimumOrMaximumPerQuadrant() */ |
1138 | | /************************************************************************/ |
1139 | | |
1140 | | /** |
1141 | | * Minimum or maximum data value (data metric), with a per-quadrant search |
1142 | | * logic. |
1143 | | */ |
1144 | | template <bool IS_MIN> |
1145 | | static CPLErr GDALGridDataMetricMinimumOrMaximumPerQuadrant( |
1146 | | const void *poOptionsIn, const double *padfX, const double *padfY, |
1147 | | const double *padfZ, double dfXPoint, double dfYPoint, double *pdfValue, |
1148 | | void *hExtraParamsIn) |
1149 | 0 | { |
1150 | 0 | const GDALGridDataMetricsOptions *const poOptions = |
1151 | 0 | static_cast<const GDALGridDataMetricsOptions *>(poOptionsIn); |
1152 | | |
1153 | | // Pre-compute search ellipse parameters. |
1154 | 0 | const double dfRadius1Square = poOptions->dfRadius1 * poOptions->dfRadius1; |
1155 | 0 | const double dfRadius2Square = poOptions->dfRadius2 * poOptions->dfRadius2; |
1156 | 0 | const double dfSearchRadius = |
1157 | 0 | std::max(poOptions->dfRadius1, poOptions->dfRadius2); |
1158 | 0 | const double dfR12Square = dfRadius1Square * dfRadius2Square; |
1159 | | |
1160 | | // const GUInt32 nMaxPoints = poOptions->nMaxPoints; |
1161 | 0 | const GUInt32 nMinPointsPerQuadrant = poOptions->nMinPointsPerQuadrant; |
1162 | 0 | const GUInt32 nMaxPointsPerQuadrant = poOptions->nMaxPointsPerQuadrant; |
1163 | |
|
1164 | 0 | GDALGridExtraParameters *psExtraParams = |
1165 | 0 | static_cast<GDALGridExtraParameters *>(hExtraParamsIn); |
1166 | 0 | const CPLQuadTree *phQuadTree = psExtraParams->hQuadTree; |
1167 | 0 | CPLAssert(phQuadTree); |
1168 | |
|
1169 | 0 | CPLRectObj sAoi; |
1170 | 0 | sAoi.minx = dfXPoint - dfSearchRadius; |
1171 | 0 | sAoi.miny = dfYPoint - dfSearchRadius; |
1172 | 0 | sAoi.maxx = dfXPoint + dfSearchRadius; |
1173 | 0 | sAoi.maxy = dfYPoint + dfSearchRadius; |
1174 | 0 | int nFeatureCount = 0; |
1175 | 0 | GDALGridPoint **papsPoints = reinterpret_cast<GDALGridPoint **>( |
1176 | 0 | CPLQuadTreeSearch(phQuadTree, &sAoi, &nFeatureCount)); |
1177 | 0 | std::multimap<double, double> oMapDistanceToZValuesPerQuadrant[4]; |
1178 | |
|
1179 | 0 | if (nFeatureCount != 0) |
1180 | 0 | { |
1181 | 0 | for (int k = 0; k < nFeatureCount; k++) |
1182 | 0 | { |
1183 | 0 | const int i = papsPoints[k]->i; |
1184 | 0 | const double dfRX = padfX[i] - dfXPoint; |
1185 | 0 | const double dfRY = padfY[i] - dfYPoint; |
1186 | 0 | const double dfRXSquare = dfRX * dfRX; |
1187 | 0 | const double dfRYSquare = dfRY * dfRY; |
1188 | |
|
1189 | 0 | if (dfRadius2Square * dfRXSquare + dfRadius1Square * dfRYSquare <= |
1190 | 0 | dfR12Square) |
1191 | 0 | { |
1192 | 0 | const int iQuadrant = |
1193 | 0 | ((dfRX >= 0) ? 1 : 0) | (((dfRY >= 0) ? 1 : 0) << 1); |
1194 | 0 | oMapDistanceToZValuesPerQuadrant[iQuadrant].insert( |
1195 | 0 | std::make_pair(dfRXSquare + dfRYSquare, padfZ[i])); |
1196 | 0 | } |
1197 | 0 | } |
1198 | 0 | } |
1199 | 0 | CPLFree(papsPoints); |
1200 | |
|
1201 | 0 | std::multimap<double, double>::iterator aoIter[] = { |
1202 | 0 | oMapDistanceToZValuesPerQuadrant[0].begin(), |
1203 | 0 | oMapDistanceToZValuesPerQuadrant[1].begin(), |
1204 | 0 | oMapDistanceToZValuesPerQuadrant[2].begin(), |
1205 | 0 | oMapDistanceToZValuesPerQuadrant[3].begin(), |
1206 | 0 | }; |
1207 | 0 | constexpr int ALL_QUADRANT_FLAGS = 1 + 2 + 4 + 8; |
1208 | | |
1209 | | // Examine all "neighbors" within the radius (sorted by distance via the |
1210 | | // multimap), and use the closest n points based on distance until the max |
1211 | | // is reached. |
1212 | | // Do that by fetching the nearest point in quadrant 0, then the nearest |
1213 | | // point in quadrant 1, 2 and 3, and starting again with the next nearest |
1214 | | // point in quarant 0, etc. |
1215 | 0 | int nQuadrantIterFinishedFlag = 0; |
1216 | 0 | GUInt32 anPerQuadrant[4] = {0}; |
1217 | 0 | double dfExtremum = IS_MIN ? std::numeric_limits<double>::max() |
1218 | 0 | : -std::numeric_limits<double>::max(); |
1219 | 0 | GUInt32 n = 0; |
1220 | 0 | for (int iQuadrant = 0; /* true */; iQuadrant = (iQuadrant + 1) % 4) |
1221 | 0 | { |
1222 | 0 | if (aoIter[iQuadrant] == |
1223 | 0 | oMapDistanceToZValuesPerQuadrant[iQuadrant].end() || |
1224 | 0 | (nMaxPointsPerQuadrant > 0 && |
1225 | 0 | anPerQuadrant[iQuadrant] >= nMaxPointsPerQuadrant)) |
1226 | 0 | { |
1227 | 0 | nQuadrantIterFinishedFlag |= 1 << iQuadrant; |
1228 | 0 | if (nQuadrantIterFinishedFlag == ALL_QUADRANT_FLAGS) |
1229 | 0 | break; |
1230 | 0 | continue; |
1231 | 0 | } |
1232 | | |
1233 | 0 | const double dfZ = aoIter[iQuadrant]->second; |
1234 | 0 | ++aoIter[iQuadrant]; |
1235 | |
|
1236 | 0 | if (IS_MIN) |
1237 | 0 | { |
1238 | 0 | if (dfExtremum > dfZ) |
1239 | 0 | dfExtremum = dfZ; |
1240 | 0 | } |
1241 | 0 | else |
1242 | 0 | { |
1243 | 0 | if (dfExtremum < dfZ) |
1244 | 0 | dfExtremum = dfZ; |
1245 | 0 | } |
1246 | 0 | n++; |
1247 | 0 | anPerQuadrant[iQuadrant]++; |
1248 | | /*if( nMaxPoints > 0 && n >= nMaxPoints ) |
1249 | | { |
1250 | | break; |
1251 | | }*/ |
1252 | 0 | } |
1253 | |
|
1254 | 0 | if (nMinPointsPerQuadrant > 0 && |
1255 | 0 | (anPerQuadrant[0] < nMinPointsPerQuadrant || |
1256 | 0 | anPerQuadrant[1] < nMinPointsPerQuadrant || |
1257 | 0 | anPerQuadrant[2] < nMinPointsPerQuadrant || |
1258 | 0 | anPerQuadrant[3] < nMinPointsPerQuadrant)) |
1259 | 0 | { |
1260 | 0 | *pdfValue = poOptions->dfNoDataValue; |
1261 | 0 | } |
1262 | 0 | else if (n < poOptions->nMinPoints || n == 0) |
1263 | 0 | { |
1264 | 0 | *pdfValue = poOptions->dfNoDataValue; |
1265 | 0 | } |
1266 | 0 | else |
1267 | 0 | { |
1268 | 0 | *pdfValue = dfExtremum; |
1269 | 0 | } |
1270 | |
|
1271 | 0 | return CE_None; |
1272 | 0 | } Unexecuted instantiation: gdalgrid.cpp:CPLErr GDALGridDataMetricMinimumOrMaximumPerQuadrant<true>(void const*, double const*, double const*, double const*, double, double, double*, void*) Unexecuted instantiation: gdalgrid.cpp:CPLErr GDALGridDataMetricMinimumOrMaximumPerQuadrant<false>(void const*, double const*, double const*, double const*, double, double, double*, void*) |
1273 | | |
1274 | | /************************************************************************/ |
1275 | | /* GDALGridDataMetricMinimumPerQuadrant() */ |
1276 | | /************************************************************************/ |
1277 | | |
1278 | | /** |
1279 | | * Minimum data value (data metric), with a per-quadrant search logic. |
1280 | | */ |
1281 | | static CPLErr GDALGridDataMetricMinimumPerQuadrant( |
1282 | | const void *poOptionsIn, GUInt32 /* nPoints */, const double *padfX, |
1283 | | const double *padfY, const double *padfZ, double dfXPoint, double dfYPoint, |
1284 | | double *pdfValue, void *hExtraParamsIn) |
1285 | 0 | { |
1286 | 0 | return GDALGridDataMetricMinimumOrMaximumPerQuadrant</*IS_MIN=*/true>( |
1287 | 0 | poOptionsIn, padfX, padfY, padfZ, dfXPoint, dfYPoint, pdfValue, |
1288 | 0 | hExtraParamsIn); |
1289 | 0 | } |
1290 | | |
1291 | | /************************************************************************/ |
1292 | | /* GDALGridDataMetricMaximum() */ |
1293 | | /************************************************************************/ |
1294 | | |
1295 | | /** |
1296 | | * Maximum data value (data metric). |
1297 | | * |
1298 | | * Maximum value found in grid node search ellipse. If there are no points |
1299 | | * found, the specified NODATA value will be returned. |
1300 | | * |
1301 | | * \f[ |
1302 | | * Z=\max{(Z_1,Z_2,\ldots,Z_n)} |
1303 | | * \f] |
1304 | | * |
1305 | | * where |
1306 | | * <ul> |
1307 | | * <li> \f$Z\f$ is a resulting value at the grid node, |
1308 | | * <li> \f$Z_i\f$ is a known value at point \f$i\f$, |
1309 | | * <li> \f$n\f$ is a total number of points in search ellipse. |
1310 | | * </ul> |
1311 | | * |
1312 | | * @param poOptionsIn Algorithm parameters. This should point to |
1313 | | * GDALGridDataMetricsOptions object. |
1314 | | * @param nPoints Number of elements in input arrays. |
1315 | | * @param padfX Input array of X coordinates. |
1316 | | * @param padfY Input array of Y coordinates. |
1317 | | * @param padfZ Input array of Z values. |
1318 | | * @param dfXPoint X coordinate of the point to compute. |
1319 | | * @param dfYPoint Y coordinate of the point to compute. |
1320 | | * @param pdfValue Pointer to variable where the computed grid node value |
1321 | | * will be returned. |
1322 | | * @param hExtraParamsIn extra parameters (unused) |
1323 | | * |
1324 | | * @return CE_None on success or CE_Failure if something goes wrong. |
1325 | | */ |
1326 | | |
1327 | | CPLErr GDALGridDataMetricMaximum(const void *poOptionsIn, GUInt32 nPoints, |
1328 | | const double *padfX, const double *padfY, |
1329 | | const double *padfZ, double dfXPoint, |
1330 | | double dfYPoint, double *pdfValue, |
1331 | | void *hExtraParamsIn) |
1332 | 0 | { |
1333 | | // TODO: For optimization purposes pre-computed parameters should be moved |
1334 | | // out of this routine to the calling function. |
1335 | |
|
1336 | 0 | const GDALGridDataMetricsOptions *const poOptions = |
1337 | 0 | static_cast<const GDALGridDataMetricsOptions *>(poOptionsIn); |
1338 | | |
1339 | | // Pre-compute search ellipse parameters. |
1340 | 0 | const double dfRadius1Square = poOptions->dfRadius1 * poOptions->dfRadius1; |
1341 | 0 | const double dfRadius2Square = poOptions->dfRadius2 * poOptions->dfRadius2; |
1342 | 0 | const double dfSearchRadius = |
1343 | 0 | std::max(poOptions->dfRadius1, poOptions->dfRadius2); |
1344 | 0 | const double dfR12Square = dfRadius1Square * dfRadius2Square; |
1345 | |
|
1346 | 0 | GDALGridExtraParameters *psExtraParams = |
1347 | 0 | static_cast<GDALGridExtraParameters *>(hExtraParamsIn); |
1348 | 0 | CPLQuadTree *phQuadTree = psExtraParams->hQuadTree; |
1349 | | |
1350 | | // Compute coefficients for coordinate system rotation. |
1351 | 0 | const double dfAngle = TO_RADIANS * poOptions->dfAngle; |
1352 | 0 | const bool bRotated = dfAngle != 0.0; |
1353 | 0 | const double dfCoeff1 = bRotated ? cos(dfAngle) : 0.0; |
1354 | 0 | const double dfCoeff2 = bRotated ? sin(dfAngle) : 0.0; |
1355 | |
|
1356 | 0 | double dfMaximumValue = -std::numeric_limits<double>::max(); |
1357 | 0 | GUInt32 n = 0; |
1358 | 0 | if (phQuadTree != nullptr) |
1359 | 0 | { |
1360 | 0 | CPLRectObj sAoi; |
1361 | 0 | sAoi.minx = dfXPoint - dfSearchRadius; |
1362 | 0 | sAoi.miny = dfYPoint - dfSearchRadius; |
1363 | 0 | sAoi.maxx = dfXPoint + dfSearchRadius; |
1364 | 0 | sAoi.maxy = dfYPoint + dfSearchRadius; |
1365 | 0 | int nFeatureCount = 0; |
1366 | 0 | GDALGridPoint **papsPoints = reinterpret_cast<GDALGridPoint **>( |
1367 | 0 | CPLQuadTreeSearch(phQuadTree, &sAoi, &nFeatureCount)); |
1368 | 0 | if (nFeatureCount != 0) |
1369 | 0 | { |
1370 | 0 | for (int k = 0; k < nFeatureCount; k++) |
1371 | 0 | { |
1372 | 0 | const int i = papsPoints[k]->i; |
1373 | 0 | const double dfRX = padfX[i] - dfXPoint; |
1374 | 0 | const double dfRY = padfY[i] - dfYPoint; |
1375 | |
|
1376 | 0 | if (dfRadius2Square * dfRX * dfRX + |
1377 | 0 | dfRadius1Square * dfRY * dfRY <= |
1378 | 0 | dfR12Square) |
1379 | 0 | { |
1380 | 0 | if (dfMaximumValue < padfZ[i]) |
1381 | 0 | dfMaximumValue = padfZ[i]; |
1382 | 0 | n++; |
1383 | 0 | } |
1384 | 0 | } |
1385 | 0 | } |
1386 | 0 | CPLFree(papsPoints); |
1387 | 0 | } |
1388 | 0 | else |
1389 | 0 | { |
1390 | 0 | GUInt32 i = 0; |
1391 | 0 | while (i < nPoints) |
1392 | 0 | { |
1393 | 0 | double dfRX = padfX[i] - dfXPoint; |
1394 | 0 | double dfRY = padfY[i] - dfYPoint; |
1395 | |
|
1396 | 0 | if (bRotated) |
1397 | 0 | { |
1398 | 0 | const double dfRXRotated = dfRX * dfCoeff1 + dfRY * dfCoeff2; |
1399 | 0 | const double dfRYRotated = dfRY * dfCoeff1 - dfRX * dfCoeff2; |
1400 | |
|
1401 | 0 | dfRX = dfRXRotated; |
1402 | 0 | dfRY = dfRYRotated; |
1403 | 0 | } |
1404 | | |
1405 | | // Is this point located inside the search ellipse? |
1406 | 0 | if (dfRadius2Square * dfRX * dfRX + dfRadius1Square * dfRY * dfRY <= |
1407 | 0 | dfR12Square) |
1408 | 0 | { |
1409 | 0 | if (dfMaximumValue < padfZ[i]) |
1410 | 0 | dfMaximumValue = padfZ[i]; |
1411 | 0 | n++; |
1412 | 0 | } |
1413 | |
|
1414 | 0 | i++; |
1415 | 0 | } |
1416 | 0 | } |
1417 | |
|
1418 | 0 | if (n < poOptions->nMinPoints || n == 0) |
1419 | 0 | { |
1420 | 0 | *pdfValue = poOptions->dfNoDataValue; |
1421 | 0 | } |
1422 | 0 | else |
1423 | 0 | { |
1424 | 0 | *pdfValue = dfMaximumValue; |
1425 | 0 | } |
1426 | |
|
1427 | 0 | return CE_None; |
1428 | 0 | } |
1429 | | |
1430 | | /************************************************************************/ |
1431 | | /* GDALGridDataMetricMaximumPerQuadrant() */ |
1432 | | /************************************************************************/ |
1433 | | |
1434 | | /** |
1435 | | * Maximum data value (data metric), with a per-quadrant search logic. |
1436 | | */ |
1437 | | static CPLErr GDALGridDataMetricMaximumPerQuadrant( |
1438 | | const void *poOptionsIn, GUInt32 /* nPoints */, const double *padfX, |
1439 | | const double *padfY, const double *padfZ, double dfXPoint, double dfYPoint, |
1440 | | double *pdfValue, void *hExtraParamsIn) |
1441 | 0 | { |
1442 | 0 | return GDALGridDataMetricMinimumOrMaximumPerQuadrant</*IS_MIN=*/false>( |
1443 | 0 | poOptionsIn, padfX, padfY, padfZ, dfXPoint, dfYPoint, pdfValue, |
1444 | 0 | hExtraParamsIn); |
1445 | 0 | } |
1446 | | |
1447 | | /************************************************************************/ |
1448 | | /* GDALGridDataMetricRange() */ |
1449 | | /************************************************************************/ |
1450 | | |
1451 | | /** |
1452 | | * Data range (data metric). |
1453 | | * |
1454 | | * A difference between the minimum and maximum values found in grid node |
1455 | | * search ellipse. If there are no points found, the specified NODATA |
1456 | | * value will be returned. |
1457 | | * |
1458 | | * \f[ |
1459 | | * Z=\max{(Z_1,Z_2,\ldots,Z_n)}-\min{(Z_1,Z_2,\ldots,Z_n)} |
1460 | | * \f] |
1461 | | * |
1462 | | * where |
1463 | | * <ul> |
1464 | | * <li> \f$Z\f$ is a resulting value at the grid node, |
1465 | | * <li> \f$Z_i\f$ is a known value at point \f$i\f$, |
1466 | | * <li> \f$n\f$ is a total number of points in search ellipse. |
1467 | | * </ul> |
1468 | | * |
1469 | | * @param poOptionsIn Algorithm parameters. This should point to |
1470 | | * GDALGridDataMetricsOptions object. |
1471 | | * @param nPoints Number of elements in input arrays. |
1472 | | * @param padfX Input array of X coordinates. |
1473 | | * @param padfY Input array of Y coordinates. |
1474 | | * @param padfZ Input array of Z values. |
1475 | | * @param dfXPoint X coordinate of the point to compute. |
1476 | | * @param dfYPoint Y coordinate of the point to compute. |
1477 | | * @param pdfValue Pointer to variable where the computed grid node value |
1478 | | * will be returned. |
1479 | | * @param hExtraParamsIn extra parameters (unused) |
1480 | | * |
1481 | | * @return CE_None on success or CE_Failure if something goes wrong. |
1482 | | */ |
1483 | | |
1484 | | CPLErr GDALGridDataMetricRange(const void *poOptionsIn, GUInt32 nPoints, |
1485 | | const double *padfX, const double *padfY, |
1486 | | const double *padfZ, double dfXPoint, |
1487 | | double dfYPoint, double *pdfValue, |
1488 | | void *hExtraParamsIn) |
1489 | 0 | { |
1490 | | // TODO: For optimization purposes pre-computed parameters should be moved |
1491 | | // out of this routine to the calling function. |
1492 | |
|
1493 | 0 | const GDALGridDataMetricsOptions *const poOptions = |
1494 | 0 | static_cast<const GDALGridDataMetricsOptions *>(poOptionsIn); |
1495 | | // Pre-compute search ellipse parameters. |
1496 | 0 | const double dfRadius1Square = poOptions->dfRadius1 * poOptions->dfRadius1; |
1497 | 0 | const double dfRadius2Square = poOptions->dfRadius2 * poOptions->dfRadius2; |
1498 | 0 | const double dfSearchRadius = |
1499 | 0 | std::max(poOptions->dfRadius1, poOptions->dfRadius2); |
1500 | 0 | const double dfR12Square = dfRadius1Square * dfRadius2Square; |
1501 | |
|
1502 | 0 | GDALGridExtraParameters *psExtraParams = |
1503 | 0 | static_cast<GDALGridExtraParameters *>(hExtraParamsIn); |
1504 | 0 | CPLQuadTree *phQuadTree = psExtraParams->hQuadTree; |
1505 | | |
1506 | | // Compute coefficients for coordinate system rotation. |
1507 | 0 | const double dfAngle = TO_RADIANS * poOptions->dfAngle; |
1508 | 0 | const bool bRotated = dfAngle != 0.0; |
1509 | 0 | const double dfCoeff1 = bRotated ? cos(dfAngle) : 0.0; |
1510 | 0 | const double dfCoeff2 = bRotated ? sin(dfAngle) : 0.0; |
1511 | |
|
1512 | 0 | double dfMaximumValue = -std::numeric_limits<double>::max(); |
1513 | 0 | double dfMinimumValue = std::numeric_limits<double>::max(); |
1514 | 0 | GUInt32 n = 0; |
1515 | 0 | if (phQuadTree != nullptr) |
1516 | 0 | { |
1517 | 0 | CPLRectObj sAoi; |
1518 | 0 | sAoi.minx = dfXPoint - dfSearchRadius; |
1519 | 0 | sAoi.miny = dfYPoint - dfSearchRadius; |
1520 | 0 | sAoi.maxx = dfXPoint + dfSearchRadius; |
1521 | 0 | sAoi.maxy = dfYPoint + dfSearchRadius; |
1522 | 0 | int nFeatureCount = 0; |
1523 | 0 | GDALGridPoint **papsPoints = reinterpret_cast<GDALGridPoint **>( |
1524 | 0 | CPLQuadTreeSearch(phQuadTree, &sAoi, &nFeatureCount)); |
1525 | 0 | if (nFeatureCount != 0) |
1526 | 0 | { |
1527 | 0 | for (int k = 0; k < nFeatureCount; k++) |
1528 | 0 | { |
1529 | 0 | const int i = papsPoints[k]->i; |
1530 | 0 | const double dfRX = padfX[i] - dfXPoint; |
1531 | 0 | const double dfRY = padfY[i] - dfYPoint; |
1532 | |
|
1533 | 0 | if (dfRadius2Square * dfRX * dfRX + |
1534 | 0 | dfRadius1Square * dfRY * dfRY <= |
1535 | 0 | dfR12Square) |
1536 | 0 | { |
1537 | 0 | if (dfMinimumValue > padfZ[i]) |
1538 | 0 | dfMinimumValue = padfZ[i]; |
1539 | 0 | if (dfMaximumValue < padfZ[i]) |
1540 | 0 | dfMaximumValue = padfZ[i]; |
1541 | 0 | n++; |
1542 | 0 | } |
1543 | 0 | } |
1544 | 0 | } |
1545 | 0 | CPLFree(papsPoints); |
1546 | 0 | } |
1547 | 0 | else |
1548 | 0 | { |
1549 | 0 | GUInt32 i = 0; |
1550 | 0 | while (i < nPoints) |
1551 | 0 | { |
1552 | 0 | double dfRX = padfX[i] - dfXPoint; |
1553 | 0 | double dfRY = padfY[i] - dfYPoint; |
1554 | |
|
1555 | 0 | if (bRotated) |
1556 | 0 | { |
1557 | 0 | const double dfRXRotated = dfRX * dfCoeff1 + dfRY * dfCoeff2; |
1558 | 0 | const double dfRYRotated = dfRY * dfCoeff1 - dfRX * dfCoeff2; |
1559 | |
|
1560 | 0 | dfRX = dfRXRotated; |
1561 | 0 | dfRY = dfRYRotated; |
1562 | 0 | } |
1563 | | |
1564 | | // Is this point located inside the search ellipse? |
1565 | 0 | if (dfRadius2Square * dfRX * dfRX + dfRadius1Square * dfRY * dfRY <= |
1566 | 0 | dfR12Square) |
1567 | 0 | { |
1568 | 0 | if (dfMinimumValue > padfZ[i]) |
1569 | 0 | dfMinimumValue = padfZ[i]; |
1570 | 0 | if (dfMaximumValue < padfZ[i]) |
1571 | 0 | dfMaximumValue = padfZ[i]; |
1572 | 0 | n++; |
1573 | 0 | } |
1574 | |
|
1575 | 0 | i++; |
1576 | 0 | } |
1577 | 0 | } |
1578 | |
|
1579 | 0 | if (n < poOptions->nMinPoints || n == 0) |
1580 | 0 | { |
1581 | 0 | *pdfValue = poOptions->dfNoDataValue; |
1582 | 0 | } |
1583 | 0 | else |
1584 | 0 | { |
1585 | 0 | *pdfValue = dfMaximumValue - dfMinimumValue; |
1586 | 0 | } |
1587 | |
|
1588 | 0 | return CE_None; |
1589 | 0 | } |
1590 | | |
1591 | | /************************************************************************/ |
1592 | | /* GDALGridDataMetricRangePerQuadrant() */ |
1593 | | /************************************************************************/ |
1594 | | |
1595 | | /** |
1596 | | * Data range (data metric), with a per-quadrant search logic. |
1597 | | */ |
1598 | | static CPLErr GDALGridDataMetricRangePerQuadrant( |
1599 | | const void *poOptionsIn, GUInt32 /* nPoints */, const double *padfX, |
1600 | | const double *padfY, const double *padfZ, double dfXPoint, double dfYPoint, |
1601 | | double *pdfValue, void *hExtraParamsIn) |
1602 | 0 | { |
1603 | 0 | const GDALGridDataMetricsOptions *const poOptions = |
1604 | 0 | static_cast<const GDALGridDataMetricsOptions *>(poOptionsIn); |
1605 | | |
1606 | | // Pre-compute search ellipse parameters. |
1607 | 0 | const double dfRadius1Square = poOptions->dfRadius1 * poOptions->dfRadius1; |
1608 | 0 | const double dfRadius2Square = poOptions->dfRadius2 * poOptions->dfRadius2; |
1609 | 0 | const double dfSearchRadius = |
1610 | 0 | std::max(poOptions->dfRadius1, poOptions->dfRadius2); |
1611 | 0 | const double dfR12Square = dfRadius1Square * dfRadius2Square; |
1612 | | |
1613 | | // const GUInt32 nMaxPoints = poOptions->nMaxPoints; |
1614 | 0 | const GUInt32 nMinPointsPerQuadrant = poOptions->nMinPointsPerQuadrant; |
1615 | 0 | const GUInt32 nMaxPointsPerQuadrant = poOptions->nMaxPointsPerQuadrant; |
1616 | |
|
1617 | 0 | GDALGridExtraParameters *psExtraParams = |
1618 | 0 | static_cast<GDALGridExtraParameters *>(hExtraParamsIn); |
1619 | 0 | const CPLQuadTree *phQuadTree = psExtraParams->hQuadTree; |
1620 | 0 | CPLAssert(phQuadTree); |
1621 | |
|
1622 | 0 | CPLRectObj sAoi; |
1623 | 0 | sAoi.minx = dfXPoint - dfSearchRadius; |
1624 | 0 | sAoi.miny = dfYPoint - dfSearchRadius; |
1625 | 0 | sAoi.maxx = dfXPoint + dfSearchRadius; |
1626 | 0 | sAoi.maxy = dfYPoint + dfSearchRadius; |
1627 | 0 | int nFeatureCount = 0; |
1628 | 0 | GDALGridPoint **papsPoints = reinterpret_cast<GDALGridPoint **>( |
1629 | 0 | CPLQuadTreeSearch(phQuadTree, &sAoi, &nFeatureCount)); |
1630 | 0 | std::multimap<double, double> oMapDistanceToZValuesPerQuadrant[4]; |
1631 | |
|
1632 | 0 | if (nFeatureCount != 0) |
1633 | 0 | { |
1634 | 0 | for (int k = 0; k < nFeatureCount; k++) |
1635 | 0 | { |
1636 | 0 | const int i = papsPoints[k]->i; |
1637 | 0 | const double dfRX = padfX[i] - dfXPoint; |
1638 | 0 | const double dfRY = padfY[i] - dfYPoint; |
1639 | 0 | const double dfRXSquare = dfRX * dfRX; |
1640 | 0 | const double dfRYSquare = dfRY * dfRY; |
1641 | |
|
1642 | 0 | if (dfRadius2Square * dfRXSquare + dfRadius1Square * dfRYSquare <= |
1643 | 0 | dfR12Square) |
1644 | 0 | { |
1645 | 0 | const int iQuadrant = |
1646 | 0 | ((dfRX >= 0) ? 1 : 0) | (((dfRY >= 0) ? 1 : 0) << 1); |
1647 | 0 | oMapDistanceToZValuesPerQuadrant[iQuadrant].insert( |
1648 | 0 | std::make_pair(dfRXSquare + dfRYSquare, padfZ[i])); |
1649 | 0 | } |
1650 | 0 | } |
1651 | 0 | } |
1652 | 0 | CPLFree(papsPoints); |
1653 | |
|
1654 | 0 | std::multimap<double, double>::iterator aoIter[] = { |
1655 | 0 | oMapDistanceToZValuesPerQuadrant[0].begin(), |
1656 | 0 | oMapDistanceToZValuesPerQuadrant[1].begin(), |
1657 | 0 | oMapDistanceToZValuesPerQuadrant[2].begin(), |
1658 | 0 | oMapDistanceToZValuesPerQuadrant[3].begin(), |
1659 | 0 | }; |
1660 | 0 | constexpr int ALL_QUADRANT_FLAGS = 1 + 2 + 4 + 8; |
1661 | | |
1662 | | // Examine all "neighbors" within the radius (sorted by distance via the |
1663 | | // multimap), and use the closest n points based on distance until the max |
1664 | | // is reached. |
1665 | | // Do that by fetching the nearest point in quadrant 0, then the nearest |
1666 | | // point in quadrant 1, 2 and 3, and starting again with the next nearest |
1667 | | // point in quarant 0, etc. |
1668 | 0 | int nQuadrantIterFinishedFlag = 0; |
1669 | 0 | GUInt32 anPerQuadrant[4] = {0}; |
1670 | 0 | double dfMaximumValue = -std::numeric_limits<double>::max(); |
1671 | 0 | double dfMinimumValue = std::numeric_limits<double>::max(); |
1672 | 0 | GUInt32 n = 0; |
1673 | 0 | for (int iQuadrant = 0; /* true */; iQuadrant = (iQuadrant + 1) % 4) |
1674 | 0 | { |
1675 | 0 | if (aoIter[iQuadrant] == |
1676 | 0 | oMapDistanceToZValuesPerQuadrant[iQuadrant].end() || |
1677 | 0 | (nMaxPointsPerQuadrant > 0 && |
1678 | 0 | anPerQuadrant[iQuadrant] >= nMaxPointsPerQuadrant)) |
1679 | 0 | { |
1680 | 0 | nQuadrantIterFinishedFlag |= 1 << iQuadrant; |
1681 | 0 | if (nQuadrantIterFinishedFlag == ALL_QUADRANT_FLAGS) |
1682 | 0 | break; |
1683 | 0 | continue; |
1684 | 0 | } |
1685 | | |
1686 | 0 | const double dfZ = aoIter[iQuadrant]->second; |
1687 | 0 | ++aoIter[iQuadrant]; |
1688 | |
|
1689 | 0 | if (dfMinimumValue > dfZ) |
1690 | 0 | dfMinimumValue = dfZ; |
1691 | 0 | if (dfMaximumValue < dfZ) |
1692 | 0 | dfMaximumValue = dfZ; |
1693 | 0 | n++; |
1694 | 0 | anPerQuadrant[iQuadrant]++; |
1695 | | /*if( nMaxPoints > 0 && n >= nMaxPoints ) |
1696 | | { |
1697 | | break; |
1698 | | }*/ |
1699 | 0 | } |
1700 | |
|
1701 | 0 | if (nMinPointsPerQuadrant > 0 && |
1702 | 0 | (anPerQuadrant[0] < nMinPointsPerQuadrant || |
1703 | 0 | anPerQuadrant[1] < nMinPointsPerQuadrant || |
1704 | 0 | anPerQuadrant[2] < nMinPointsPerQuadrant || |
1705 | 0 | anPerQuadrant[3] < nMinPointsPerQuadrant)) |
1706 | 0 | { |
1707 | 0 | *pdfValue = poOptions->dfNoDataValue; |
1708 | 0 | } |
1709 | 0 | else if (n < poOptions->nMinPoints || n == 0) |
1710 | 0 | { |
1711 | 0 | *pdfValue = poOptions->dfNoDataValue; |
1712 | 0 | } |
1713 | 0 | else |
1714 | 0 | { |
1715 | 0 | *pdfValue = dfMaximumValue - dfMinimumValue; |
1716 | 0 | } |
1717 | |
|
1718 | 0 | return CE_None; |
1719 | 0 | } |
1720 | | |
1721 | | /************************************************************************/ |
1722 | | /* GDALGridDataMetricCount() */ |
1723 | | /************************************************************************/ |
1724 | | |
1725 | | /** |
1726 | | * Number of data points (data metric). |
1727 | | * |
1728 | | * A number of data points found in grid node search ellipse. |
1729 | | * |
1730 | | * \f[ |
1731 | | * Z=n |
1732 | | * \f] |
1733 | | * |
1734 | | * where |
1735 | | * <ul> |
1736 | | * <li> \f$Z\f$ is a resulting value at the grid node, |
1737 | | * <li> \f$n\f$ is a total number of points in search ellipse. |
1738 | | * </ul> |
1739 | | * |
1740 | | * @param poOptionsIn Algorithm parameters. This should point to |
1741 | | * GDALGridDataMetricsOptions object. |
1742 | | * @param nPoints Number of elements in input arrays. |
1743 | | * @param padfX Input array of X coordinates. |
1744 | | * @param padfY Input array of Y coordinates. |
1745 | | * @param padfZ Input array of Z values. |
1746 | | * @param dfXPoint X coordinate of the point to compute. |
1747 | | * @param dfYPoint Y coordinate of the point to compute. |
1748 | | * @param pdfValue Pointer to variable where the computed grid node value |
1749 | | * will be returned. |
1750 | | * @param hExtraParamsIn extra parameters (unused) |
1751 | | * |
1752 | | * @return CE_None on success or CE_Failure if something goes wrong. |
1753 | | */ |
1754 | | |
1755 | | CPLErr GDALGridDataMetricCount(const void *poOptionsIn, GUInt32 nPoints, |
1756 | | const double *padfX, const double *padfY, |
1757 | | CPL_UNUSED const double *padfZ, double dfXPoint, |
1758 | | double dfYPoint, double *pdfValue, |
1759 | | void *hExtraParamsIn) |
1760 | 0 | { |
1761 | | // TODO: For optimization purposes pre-computed parameters should be moved |
1762 | | // out of this routine to the calling function. |
1763 | |
|
1764 | 0 | const GDALGridDataMetricsOptions *const poOptions = |
1765 | 0 | static_cast<const GDALGridDataMetricsOptions *>(poOptionsIn); |
1766 | | |
1767 | | // Pre-compute search ellipse parameters. |
1768 | 0 | const double dfRadius1Square = poOptions->dfRadius1 * poOptions->dfRadius1; |
1769 | 0 | const double dfRadius2Square = poOptions->dfRadius2 * poOptions->dfRadius2; |
1770 | 0 | const double dfSearchRadius = |
1771 | 0 | std::max(poOptions->dfRadius1, poOptions->dfRadius2); |
1772 | 0 | const double dfR12Square = dfRadius1Square * dfRadius2Square; |
1773 | |
|
1774 | 0 | GDALGridExtraParameters *psExtraParams = |
1775 | 0 | static_cast<GDALGridExtraParameters *>(hExtraParamsIn); |
1776 | 0 | CPLQuadTree *phQuadTree = psExtraParams->hQuadTree; |
1777 | | |
1778 | | // Compute coefficients for coordinate system rotation. |
1779 | 0 | const double dfAngle = TO_RADIANS * poOptions->dfAngle; |
1780 | 0 | const bool bRotated = dfAngle != 0.0; |
1781 | 0 | const double dfCoeff1 = bRotated ? cos(dfAngle) : 0.0; |
1782 | 0 | const double dfCoeff2 = bRotated ? sin(dfAngle) : 0.0; |
1783 | |
|
1784 | 0 | GUInt32 n = 0; |
1785 | 0 | if (phQuadTree != nullptr) |
1786 | 0 | { |
1787 | 0 | CPLRectObj sAoi; |
1788 | 0 | sAoi.minx = dfXPoint - dfSearchRadius; |
1789 | 0 | sAoi.miny = dfYPoint - dfSearchRadius; |
1790 | 0 | sAoi.maxx = dfXPoint + dfSearchRadius; |
1791 | 0 | sAoi.maxy = dfYPoint + dfSearchRadius; |
1792 | 0 | int nFeatureCount = 0; |
1793 | 0 | GDALGridPoint **papsPoints = reinterpret_cast<GDALGridPoint **>( |
1794 | 0 | CPLQuadTreeSearch(phQuadTree, &sAoi, &nFeatureCount)); |
1795 | 0 | if (nFeatureCount != 0) |
1796 | 0 | { |
1797 | 0 | for (int k = 0; k < nFeatureCount; k++) |
1798 | 0 | { |
1799 | 0 | const int i = papsPoints[k]->i; |
1800 | 0 | const double dfRX = padfX[i] - dfXPoint; |
1801 | 0 | const double dfRY = padfY[i] - dfYPoint; |
1802 | |
|
1803 | 0 | if (dfRadius2Square * dfRX * dfRX + |
1804 | 0 | dfRadius1Square * dfRY * dfRY <= |
1805 | 0 | dfR12Square) |
1806 | 0 | { |
1807 | 0 | n++; |
1808 | 0 | } |
1809 | 0 | } |
1810 | 0 | } |
1811 | 0 | CPLFree(papsPoints); |
1812 | 0 | } |
1813 | 0 | else |
1814 | 0 | { |
1815 | 0 | GUInt32 i = 0; |
1816 | 0 | while (i < nPoints) |
1817 | 0 | { |
1818 | 0 | double dfRX = padfX[i] - dfXPoint; |
1819 | 0 | double dfRY = padfY[i] - dfYPoint; |
1820 | |
|
1821 | 0 | if (bRotated) |
1822 | 0 | { |
1823 | 0 | const double dfRXRotated = dfRX * dfCoeff1 + dfRY * dfCoeff2; |
1824 | 0 | const double dfRYRotated = dfRY * dfCoeff1 - dfRX * dfCoeff2; |
1825 | |
|
1826 | 0 | dfRX = dfRXRotated; |
1827 | 0 | dfRY = dfRYRotated; |
1828 | 0 | } |
1829 | | |
1830 | | // Is this point located inside the search ellipse? |
1831 | 0 | if (dfRadius2Square * dfRX * dfRX + dfRadius1Square * dfRY * dfRY <= |
1832 | 0 | dfR12Square) |
1833 | 0 | { |
1834 | 0 | n++; |
1835 | 0 | } |
1836 | |
|
1837 | 0 | i++; |
1838 | 0 | } |
1839 | 0 | } |
1840 | |
|
1841 | 0 | if (n < poOptions->nMinPoints) |
1842 | 0 | { |
1843 | 0 | *pdfValue = poOptions->dfNoDataValue; |
1844 | 0 | } |
1845 | 0 | else |
1846 | 0 | { |
1847 | 0 | *pdfValue = static_cast<double>(n); |
1848 | 0 | } |
1849 | |
|
1850 | 0 | return CE_None; |
1851 | 0 | } |
1852 | | |
1853 | | /************************************************************************/ |
1854 | | /* GDALGridDataMetricCountPerQuadrant() */ |
1855 | | /************************************************************************/ |
1856 | | |
1857 | | /** |
1858 | | * Number of data points (data metric), with a per-quadrant search logic. |
1859 | | */ |
1860 | | static CPLErr GDALGridDataMetricCountPerQuadrant( |
1861 | | const void *poOptionsIn, GUInt32 /* nPoints */, const double *padfX, |
1862 | | const double *padfY, const double *padfZ, double dfXPoint, double dfYPoint, |
1863 | | double *pdfValue, void *hExtraParamsIn) |
1864 | 0 | { |
1865 | 0 | const GDALGridDataMetricsOptions *const poOptions = |
1866 | 0 | static_cast<const GDALGridDataMetricsOptions *>(poOptionsIn); |
1867 | | |
1868 | | // Pre-compute search ellipse parameters. |
1869 | 0 | const double dfRadius1Square = poOptions->dfRadius1 * poOptions->dfRadius1; |
1870 | 0 | const double dfRadius2Square = poOptions->dfRadius2 * poOptions->dfRadius2; |
1871 | 0 | const double dfSearchRadius = |
1872 | 0 | std::max(poOptions->dfRadius1, poOptions->dfRadius2); |
1873 | 0 | const double dfR12Square = dfRadius1Square * dfRadius2Square; |
1874 | | |
1875 | | // const GUInt32 nMaxPoints = poOptions->nMaxPoints; |
1876 | 0 | const GUInt32 nMinPointsPerQuadrant = poOptions->nMinPointsPerQuadrant; |
1877 | 0 | const GUInt32 nMaxPointsPerQuadrant = poOptions->nMaxPointsPerQuadrant; |
1878 | |
|
1879 | 0 | GDALGridExtraParameters *psExtraParams = |
1880 | 0 | static_cast<GDALGridExtraParameters *>(hExtraParamsIn); |
1881 | 0 | const CPLQuadTree *phQuadTree = psExtraParams->hQuadTree; |
1882 | 0 | CPLAssert(phQuadTree); |
1883 | |
|
1884 | 0 | CPLRectObj sAoi; |
1885 | 0 | sAoi.minx = dfXPoint - dfSearchRadius; |
1886 | 0 | sAoi.miny = dfYPoint - dfSearchRadius; |
1887 | 0 | sAoi.maxx = dfXPoint + dfSearchRadius; |
1888 | 0 | sAoi.maxy = dfYPoint + dfSearchRadius; |
1889 | 0 | int nFeatureCount = 0; |
1890 | 0 | GDALGridPoint **papsPoints = reinterpret_cast<GDALGridPoint **>( |
1891 | 0 | CPLQuadTreeSearch(phQuadTree, &sAoi, &nFeatureCount)); |
1892 | 0 | std::multimap<double, double> oMapDistanceToZValuesPerQuadrant[4]; |
1893 | |
|
1894 | 0 | if (nFeatureCount != 0) |
1895 | 0 | { |
1896 | 0 | for (int k = 0; k < nFeatureCount; k++) |
1897 | 0 | { |
1898 | 0 | const int i = papsPoints[k]->i; |
1899 | 0 | const double dfRX = padfX[i] - dfXPoint; |
1900 | 0 | const double dfRY = padfY[i] - dfYPoint; |
1901 | 0 | const double dfRXSquare = dfRX * dfRX; |
1902 | 0 | const double dfRYSquare = dfRY * dfRY; |
1903 | |
|
1904 | 0 | if (dfRadius2Square * dfRXSquare + dfRadius1Square * dfRYSquare <= |
1905 | 0 | dfR12Square) |
1906 | 0 | { |
1907 | 0 | const int iQuadrant = |
1908 | 0 | ((dfRX >= 0) ? 1 : 0) | (((dfRY >= 0) ? 1 : 0) << 1); |
1909 | 0 | oMapDistanceToZValuesPerQuadrant[iQuadrant].insert( |
1910 | 0 | std::make_pair(dfRXSquare + dfRYSquare, padfZ[i])); |
1911 | 0 | } |
1912 | 0 | } |
1913 | 0 | } |
1914 | 0 | CPLFree(papsPoints); |
1915 | |
|
1916 | 0 | std::multimap<double, double>::iterator aoIter[] = { |
1917 | 0 | oMapDistanceToZValuesPerQuadrant[0].begin(), |
1918 | 0 | oMapDistanceToZValuesPerQuadrant[1].begin(), |
1919 | 0 | oMapDistanceToZValuesPerQuadrant[2].begin(), |
1920 | 0 | oMapDistanceToZValuesPerQuadrant[3].begin(), |
1921 | 0 | }; |
1922 | 0 | constexpr int ALL_QUADRANT_FLAGS = 1 + 2 + 4 + 8; |
1923 | | |
1924 | | // Examine all "neighbors" within the radius (sorted by distance via the |
1925 | | // multimap), and use the closest n points based on distance until the max |
1926 | | // is reached. |
1927 | | // Do that by fetching the nearest point in quadrant 0, then the nearest |
1928 | | // point in quadrant 1, 2 and 3, and starting again with the next nearest |
1929 | | // point in quarant 0, etc. |
1930 | 0 | int nQuadrantIterFinishedFlag = 0; |
1931 | 0 | GUInt32 anPerQuadrant[4] = {0}; |
1932 | 0 | GUInt32 n = 0; |
1933 | 0 | for (int iQuadrant = 0; /* true */; iQuadrant = (iQuadrant + 1) % 4) |
1934 | 0 | { |
1935 | 0 | if (aoIter[iQuadrant] == |
1936 | 0 | oMapDistanceToZValuesPerQuadrant[iQuadrant].end() || |
1937 | 0 | (nMaxPointsPerQuadrant > 0 && |
1938 | 0 | anPerQuadrant[iQuadrant] >= nMaxPointsPerQuadrant)) |
1939 | 0 | { |
1940 | 0 | nQuadrantIterFinishedFlag |= 1 << iQuadrant; |
1941 | 0 | if (nQuadrantIterFinishedFlag == ALL_QUADRANT_FLAGS) |
1942 | 0 | break; |
1943 | 0 | continue; |
1944 | 0 | } |
1945 | | |
1946 | 0 | ++aoIter[iQuadrant]; |
1947 | |
|
1948 | 0 | n++; |
1949 | 0 | anPerQuadrant[iQuadrant]++; |
1950 | | /*if( nMaxPoints > 0 && n >= nMaxPoints ) |
1951 | | { |
1952 | | break; |
1953 | | }*/ |
1954 | 0 | } |
1955 | |
|
1956 | 0 | if (nMinPointsPerQuadrant > 0 && |
1957 | 0 | (anPerQuadrant[0] < nMinPointsPerQuadrant || |
1958 | 0 | anPerQuadrant[1] < nMinPointsPerQuadrant || |
1959 | 0 | anPerQuadrant[2] < nMinPointsPerQuadrant || |
1960 | 0 | anPerQuadrant[3] < nMinPointsPerQuadrant)) |
1961 | 0 | { |
1962 | 0 | *pdfValue = poOptions->dfNoDataValue; |
1963 | 0 | } |
1964 | 0 | else if (n < poOptions->nMinPoints) |
1965 | 0 | { |
1966 | 0 | *pdfValue = poOptions->dfNoDataValue; |
1967 | 0 | } |
1968 | 0 | else |
1969 | 0 | { |
1970 | 0 | *pdfValue = static_cast<double>(n); |
1971 | 0 | } |
1972 | |
|
1973 | 0 | return CE_None; |
1974 | 0 | } |
1975 | | |
1976 | | /************************************************************************/ |
1977 | | /* GDALGridDataMetricAverageDistance() */ |
1978 | | /************************************************************************/ |
1979 | | |
1980 | | /** |
1981 | | * Average distance (data metric). |
1982 | | * |
1983 | | * An average distance between the grid node (center of the search ellipse) |
1984 | | * and all of the data points found in grid node search ellipse. If there are |
1985 | | * no points found, the specified NODATA value will be returned. |
1986 | | * |
1987 | | * \f[ |
1988 | | * Z=\frac{\sum_{i = 1}^n r_i}{n} |
1989 | | * \f] |
1990 | | * |
1991 | | * where |
1992 | | * <ul> |
1993 | | * <li> \f$Z\f$ is a resulting value at the grid node, |
1994 | | * <li> \f$r_i\f$ is an Euclidean distance from the grid node |
1995 | | * to point \f$i\f$, |
1996 | | * <li> \f$n\f$ is a total number of points in search ellipse. |
1997 | | * </ul> |
1998 | | * |
1999 | | * @param poOptionsIn Algorithm parameters. This should point to |
2000 | | * GDALGridDataMetricsOptions object. |
2001 | | * @param nPoints Number of elements in input arrays. |
2002 | | * @param padfX Input array of X coordinates. |
2003 | | * @param padfY Input array of Y coordinates. |
2004 | | * @param padfZ Input array of Z values (unused) |
2005 | | * @param dfXPoint X coordinate of the point to compute. |
2006 | | * @param dfYPoint Y coordinate of the point to compute. |
2007 | | * @param pdfValue Pointer to variable where the computed grid node value |
2008 | | * will be returned. |
2009 | | * @param hExtraParamsIn extra parameters (unused) |
2010 | | * |
2011 | | * @return CE_None on success or CE_Failure if something goes wrong. |
2012 | | */ |
2013 | | |
2014 | | CPLErr GDALGridDataMetricAverageDistance(const void *poOptionsIn, |
2015 | | GUInt32 nPoints, const double *padfX, |
2016 | | const double *padfY, |
2017 | | CPL_UNUSED const double *padfZ, |
2018 | | double dfXPoint, double dfYPoint, |
2019 | | double *pdfValue, void *hExtraParamsIn) |
2020 | 0 | { |
2021 | | // TODO: For optimization purposes pre-computed parameters should be moved |
2022 | | // out of this routine to the calling function. |
2023 | |
|
2024 | 0 | const GDALGridDataMetricsOptions *const poOptions = |
2025 | 0 | static_cast<const GDALGridDataMetricsOptions *>(poOptionsIn); |
2026 | | |
2027 | | // Pre-compute search ellipse parameters. |
2028 | 0 | const double dfRadius1Square = poOptions->dfRadius1 * poOptions->dfRadius1; |
2029 | 0 | const double dfRadius2Square = poOptions->dfRadius2 * poOptions->dfRadius2; |
2030 | 0 | const double dfSearchRadius = |
2031 | 0 | std::max(poOptions->dfRadius1, poOptions->dfRadius2); |
2032 | 0 | const double dfR12Square = dfRadius1Square * dfRadius2Square; |
2033 | |
|
2034 | 0 | GDALGridExtraParameters *psExtraParams = |
2035 | 0 | static_cast<GDALGridExtraParameters *>(hExtraParamsIn); |
2036 | 0 | CPLQuadTree *phQuadTree = psExtraParams->hQuadTree; |
2037 | | |
2038 | | // Compute coefficients for coordinate system rotation. |
2039 | 0 | const double dfAngle = TO_RADIANS * poOptions->dfAngle; |
2040 | 0 | const bool bRotated = dfAngle != 0.0; |
2041 | 0 | const double dfCoeff1 = bRotated ? cos(dfAngle) : 0.0; |
2042 | 0 | const double dfCoeff2 = bRotated ? sin(dfAngle) : 0.0; |
2043 | |
|
2044 | 0 | double dfAccumulator = 0.0; |
2045 | 0 | GUInt32 n = 0; |
2046 | 0 | if (phQuadTree != nullptr) |
2047 | 0 | { |
2048 | 0 | CPLRectObj sAoi; |
2049 | 0 | sAoi.minx = dfXPoint - dfSearchRadius; |
2050 | 0 | sAoi.miny = dfYPoint - dfSearchRadius; |
2051 | 0 | sAoi.maxx = dfXPoint + dfSearchRadius; |
2052 | 0 | sAoi.maxy = dfYPoint + dfSearchRadius; |
2053 | 0 | int nFeatureCount = 0; |
2054 | 0 | GDALGridPoint **papsPoints = reinterpret_cast<GDALGridPoint **>( |
2055 | 0 | CPLQuadTreeSearch(phQuadTree, &sAoi, &nFeatureCount)); |
2056 | 0 | if (nFeatureCount != 0) |
2057 | 0 | { |
2058 | 0 | for (int k = 0; k < nFeatureCount; k++) |
2059 | 0 | { |
2060 | 0 | const int i = papsPoints[k]->i; |
2061 | 0 | const double dfRX = padfX[i] - dfXPoint; |
2062 | 0 | const double dfRY = padfY[i] - dfYPoint; |
2063 | |
|
2064 | 0 | if (dfRadius2Square * dfRX * dfRX + |
2065 | 0 | dfRadius1Square * dfRY * dfRY <= |
2066 | 0 | dfR12Square) |
2067 | 0 | { |
2068 | 0 | dfAccumulator += sqrt(dfRX * dfRX + dfRY * dfRY); |
2069 | 0 | n++; |
2070 | 0 | } |
2071 | 0 | } |
2072 | 0 | } |
2073 | 0 | CPLFree(papsPoints); |
2074 | 0 | } |
2075 | 0 | else |
2076 | 0 | { |
2077 | 0 | GUInt32 i = 0; |
2078 | |
|
2079 | 0 | while (i < nPoints) |
2080 | 0 | { |
2081 | 0 | double dfRX = padfX[i] - dfXPoint; |
2082 | 0 | double dfRY = padfY[i] - dfYPoint; |
2083 | |
|
2084 | 0 | if (bRotated) |
2085 | 0 | { |
2086 | 0 | const double dfRXRotated = dfRX * dfCoeff1 + dfRY * dfCoeff2; |
2087 | 0 | const double dfRYRotated = dfRY * dfCoeff1 - dfRX * dfCoeff2; |
2088 | |
|
2089 | 0 | dfRX = dfRXRotated; |
2090 | 0 | dfRY = dfRYRotated; |
2091 | 0 | } |
2092 | | |
2093 | | // Is this point located inside the search ellipse? |
2094 | 0 | if (dfRadius2Square * dfRX * dfRX + dfRadius1Square * dfRY * dfRY <= |
2095 | 0 | dfR12Square) |
2096 | 0 | { |
2097 | 0 | dfAccumulator += sqrt(dfRX * dfRX + dfRY * dfRY); |
2098 | 0 | n++; |
2099 | 0 | } |
2100 | |
|
2101 | 0 | i++; |
2102 | 0 | } |
2103 | 0 | } |
2104 | |
|
2105 | 0 | if (n < poOptions->nMinPoints || n == 0) |
2106 | 0 | { |
2107 | 0 | *pdfValue = poOptions->dfNoDataValue; |
2108 | 0 | } |
2109 | 0 | else |
2110 | 0 | { |
2111 | 0 | *pdfValue = dfAccumulator / n; |
2112 | 0 | } |
2113 | |
|
2114 | 0 | return CE_None; |
2115 | 0 | } |
2116 | | |
2117 | | /************************************************************************/ |
2118 | | /* GDALGridDataMetricAverageDistancePerQuadrant() */ |
2119 | | /************************************************************************/ |
2120 | | |
2121 | | /** |
2122 | | * Average distance (data metric), with a per-quadrant search logic. |
2123 | | */ |
2124 | | static CPLErr GDALGridDataMetricAverageDistancePerQuadrant( |
2125 | | const void *poOptionsIn, GUInt32 /* nPoints */, const double *padfX, |
2126 | | const double *padfY, const double *padfZ, double dfXPoint, double dfYPoint, |
2127 | | double *pdfValue, void *hExtraParamsIn) |
2128 | 0 | { |
2129 | 0 | const GDALGridDataMetricsOptions *const poOptions = |
2130 | 0 | static_cast<const GDALGridDataMetricsOptions *>(poOptionsIn); |
2131 | | |
2132 | | // Pre-compute search ellipse parameters. |
2133 | 0 | const double dfRadius1Square = poOptions->dfRadius1 * poOptions->dfRadius1; |
2134 | 0 | const double dfRadius2Square = poOptions->dfRadius2 * poOptions->dfRadius2; |
2135 | 0 | const double dfSearchRadius = |
2136 | 0 | std::max(poOptions->dfRadius1, poOptions->dfRadius2); |
2137 | 0 | const double dfR12Square = dfRadius1Square * dfRadius2Square; |
2138 | | |
2139 | | // const GUInt32 nMaxPoints = poOptions->nMaxPoints; |
2140 | 0 | const GUInt32 nMinPointsPerQuadrant = poOptions->nMinPointsPerQuadrant; |
2141 | 0 | const GUInt32 nMaxPointsPerQuadrant = poOptions->nMaxPointsPerQuadrant; |
2142 | |
|
2143 | 0 | GDALGridExtraParameters *psExtraParams = |
2144 | 0 | static_cast<GDALGridExtraParameters *>(hExtraParamsIn); |
2145 | 0 | const CPLQuadTree *phQuadTree = psExtraParams->hQuadTree; |
2146 | 0 | CPLAssert(phQuadTree); |
2147 | |
|
2148 | 0 | CPLRectObj sAoi; |
2149 | 0 | sAoi.minx = dfXPoint - dfSearchRadius; |
2150 | 0 | sAoi.miny = dfYPoint - dfSearchRadius; |
2151 | 0 | sAoi.maxx = dfXPoint + dfSearchRadius; |
2152 | 0 | sAoi.maxy = dfYPoint + dfSearchRadius; |
2153 | 0 | int nFeatureCount = 0; |
2154 | 0 | GDALGridPoint **papsPoints = reinterpret_cast<GDALGridPoint **>( |
2155 | 0 | CPLQuadTreeSearch(phQuadTree, &sAoi, &nFeatureCount)); |
2156 | 0 | std::multimap<double, double> oMapDistanceToZValuesPerQuadrant[4]; |
2157 | |
|
2158 | 0 | if (nFeatureCount != 0) |
2159 | 0 | { |
2160 | 0 | for (int k = 0; k < nFeatureCount; k++) |
2161 | 0 | { |
2162 | 0 | const int i = papsPoints[k]->i; |
2163 | 0 | const double dfRX = padfX[i] - dfXPoint; |
2164 | 0 | const double dfRY = padfY[i] - dfYPoint; |
2165 | 0 | const double dfRXSquare = dfRX * dfRX; |
2166 | 0 | const double dfRYSquare = dfRY * dfRY; |
2167 | |
|
2168 | 0 | if (dfRadius2Square * dfRXSquare + dfRadius1Square * dfRYSquare <= |
2169 | 0 | dfR12Square) |
2170 | 0 | { |
2171 | 0 | const int iQuadrant = |
2172 | 0 | ((dfRX >= 0) ? 1 : 0) | (((dfRY >= 0) ? 1 : 0) << 1); |
2173 | 0 | oMapDistanceToZValuesPerQuadrant[iQuadrant].insert( |
2174 | 0 | std::make_pair(dfRXSquare + dfRYSquare, padfZ[i])); |
2175 | 0 | } |
2176 | 0 | } |
2177 | 0 | } |
2178 | 0 | CPLFree(papsPoints); |
2179 | |
|
2180 | 0 | std::multimap<double, double>::iterator aoIter[] = { |
2181 | 0 | oMapDistanceToZValuesPerQuadrant[0].begin(), |
2182 | 0 | oMapDistanceToZValuesPerQuadrant[1].begin(), |
2183 | 0 | oMapDistanceToZValuesPerQuadrant[2].begin(), |
2184 | 0 | oMapDistanceToZValuesPerQuadrant[3].begin(), |
2185 | 0 | }; |
2186 | 0 | constexpr int ALL_QUADRANT_FLAGS = 1 + 2 + 4 + 8; |
2187 | | |
2188 | | // Examine all "neighbors" within the radius (sorted by distance via the |
2189 | | // multimap), and use the closest n points based on distance until the max |
2190 | | // is reached. |
2191 | | // Do that by fetching the nearest point in quadrant 0, then the nearest |
2192 | | // point in quadrant 1, 2 and 3, and starting again with the next nearest |
2193 | | // point in quarant 0, etc. |
2194 | 0 | int nQuadrantIterFinishedFlag = 0; |
2195 | 0 | GUInt32 anPerQuadrant[4] = {0}; |
2196 | 0 | GUInt32 n = 0; |
2197 | 0 | double dfAccumulator = 0; |
2198 | 0 | for (int iQuadrant = 0; /* true */; iQuadrant = (iQuadrant + 1) % 4) |
2199 | 0 | { |
2200 | 0 | if (aoIter[iQuadrant] == |
2201 | 0 | oMapDistanceToZValuesPerQuadrant[iQuadrant].end() || |
2202 | 0 | (nMaxPointsPerQuadrant > 0 && |
2203 | 0 | anPerQuadrant[iQuadrant] >= nMaxPointsPerQuadrant)) |
2204 | 0 | { |
2205 | 0 | nQuadrantIterFinishedFlag |= 1 << iQuadrant; |
2206 | 0 | if (nQuadrantIterFinishedFlag == ALL_QUADRANT_FLAGS) |
2207 | 0 | break; |
2208 | 0 | continue; |
2209 | 0 | } |
2210 | | |
2211 | 0 | dfAccumulator += sqrt(aoIter[iQuadrant]->first); |
2212 | 0 | ++aoIter[iQuadrant]; |
2213 | |
|
2214 | 0 | n++; |
2215 | 0 | anPerQuadrant[iQuadrant]++; |
2216 | | /*if( nMaxPoints > 0 && n >= nMaxPoints ) |
2217 | | { |
2218 | | break; |
2219 | | }*/ |
2220 | 0 | } |
2221 | |
|
2222 | 0 | if (nMinPointsPerQuadrant > 0 && |
2223 | 0 | (anPerQuadrant[0] < nMinPointsPerQuadrant || |
2224 | 0 | anPerQuadrant[1] < nMinPointsPerQuadrant || |
2225 | 0 | anPerQuadrant[2] < nMinPointsPerQuadrant || |
2226 | 0 | anPerQuadrant[3] < nMinPointsPerQuadrant)) |
2227 | 0 | { |
2228 | 0 | *pdfValue = poOptions->dfNoDataValue; |
2229 | 0 | } |
2230 | 0 | else if (n < poOptions->nMinPoints || n == 0) |
2231 | 0 | { |
2232 | 0 | *pdfValue = poOptions->dfNoDataValue; |
2233 | 0 | } |
2234 | 0 | else |
2235 | 0 | { |
2236 | 0 | *pdfValue = dfAccumulator / n; |
2237 | 0 | } |
2238 | |
|
2239 | 0 | return CE_None; |
2240 | 0 | } |
2241 | | |
2242 | | /************************************************************************/ |
2243 | | /* GDALGridDataMetricAverageDistancePts() */ |
2244 | | /************************************************************************/ |
2245 | | |
2246 | | /** |
2247 | | * Average distance between points (data metric). |
2248 | | * |
2249 | | * An average distance between the data points found in grid node search |
2250 | | * ellipse. The distance between each pair of points within ellipse is |
2251 | | * calculated and average of all distances is set as a grid node value. If |
2252 | | * there are no points found, the specified NODATA value will be returned. |
2253 | | |
2254 | | * |
2255 | | * \f[ |
2256 | | * Z=\frac{\sum_{i = 1}^{n-1}\sum_{j=i+1}^{n} |
2257 | | r_{ij}}{\left(n-1\right)\,n-\frac{n+{\left(n-1\right)}^{2}-1}{2}} |
2258 | | * \f] |
2259 | | * |
2260 | | * where |
2261 | | * <ul> |
2262 | | * <li> \f$Z\f$ is a resulting value at the grid node, |
2263 | | * <li> \f$r_{ij}\f$ is an Euclidean distance between points |
2264 | | * \f$i\f$ and \f$j\f$, |
2265 | | * <li> \f$n\f$ is a total number of points in search ellipse. |
2266 | | * </ul> |
2267 | | * |
2268 | | * @param poOptionsIn Algorithm parameters. This should point to |
2269 | | * GDALGridDataMetricsOptions object. |
2270 | | * @param nPoints Number of elements in input arrays. |
2271 | | * @param padfX Input array of X coordinates. |
2272 | | * @param padfY Input array of Y coordinates. |
2273 | | * @param padfZ Input array of Z values (unused) |
2274 | | * @param dfXPoint X coordinate of the point to compute. |
2275 | | * @param dfYPoint Y coordinate of the point to compute. |
2276 | | * @param pdfValue Pointer to variable where the computed grid node value |
2277 | | * will be returned. |
2278 | | * @param hExtraParamsIn extra parameters (unused) |
2279 | | * |
2280 | | * @return CE_None on success or CE_Failure if something goes wrong. |
2281 | | */ |
2282 | | |
2283 | | CPLErr GDALGridDataMetricAverageDistancePts( |
2284 | | const void *poOptionsIn, GUInt32 nPoints, const double *padfX, |
2285 | | const double *padfY, CPL_UNUSED const double *padfZ, double dfXPoint, |
2286 | | double dfYPoint, double *pdfValue, void *hExtraParamsIn) |
2287 | 0 | { |
2288 | | // TODO: For optimization purposes pre-computed parameters should be moved |
2289 | | // out of this routine to the calling function. |
2290 | |
|
2291 | 0 | const GDALGridDataMetricsOptions *const poOptions = |
2292 | 0 | static_cast<const GDALGridDataMetricsOptions *>(poOptionsIn); |
2293 | | // Pre-compute search ellipse parameters. |
2294 | 0 | const double dfRadius1Square = poOptions->dfRadius1 * poOptions->dfRadius1; |
2295 | 0 | const double dfRadius2Square = poOptions->dfRadius2 * poOptions->dfRadius2; |
2296 | 0 | const double dfSearchRadius = |
2297 | 0 | std::max(poOptions->dfRadius1, poOptions->dfRadius2); |
2298 | 0 | const double dfR12Square = dfRadius1Square * dfRadius2Square; |
2299 | |
|
2300 | 0 | GDALGridExtraParameters *psExtraParams = |
2301 | 0 | static_cast<GDALGridExtraParameters *>(hExtraParamsIn); |
2302 | 0 | CPLQuadTree *phQuadTree = psExtraParams->hQuadTree; |
2303 | | |
2304 | | // Compute coefficients for coordinate system rotation. |
2305 | 0 | const double dfAngle = TO_RADIANS * poOptions->dfAngle; |
2306 | 0 | const bool bRotated = dfAngle != 0.0; |
2307 | 0 | const double dfCoeff1 = bRotated ? cos(dfAngle) : 0.0; |
2308 | 0 | const double dfCoeff2 = bRotated ? sin(dfAngle) : 0.0; |
2309 | |
|
2310 | 0 | double dfAccumulator = 0.0; |
2311 | 0 | GUInt32 n = 0; |
2312 | 0 | if (phQuadTree != nullptr) |
2313 | 0 | { |
2314 | 0 | CPLRectObj sAoi; |
2315 | 0 | sAoi.minx = dfXPoint - dfSearchRadius; |
2316 | 0 | sAoi.miny = dfYPoint - dfSearchRadius; |
2317 | 0 | sAoi.maxx = dfXPoint + dfSearchRadius; |
2318 | 0 | sAoi.maxy = dfYPoint + dfSearchRadius; |
2319 | 0 | int nFeatureCount = 0; |
2320 | 0 | GDALGridPoint **papsPoints = reinterpret_cast<GDALGridPoint **>( |
2321 | 0 | CPLQuadTreeSearch(phQuadTree, &sAoi, &nFeatureCount)); |
2322 | 0 | if (nFeatureCount != 0) |
2323 | 0 | { |
2324 | 0 | for (int k = 0; k < nFeatureCount - 1; k++) |
2325 | 0 | { |
2326 | 0 | const int i = papsPoints[k]->i; |
2327 | 0 | const double dfRX1 = padfX[i] - dfXPoint; |
2328 | 0 | const double dfRY1 = padfY[i] - dfYPoint; |
2329 | |
|
2330 | 0 | if (dfRadius2Square * dfRX1 * dfRX1 + |
2331 | 0 | dfRadius1Square * dfRY1 * dfRY1 <= |
2332 | 0 | dfR12Square) |
2333 | 0 | { |
2334 | 0 | for (int j = k; j < nFeatureCount; j++) |
2335 | | // Search all the remaining points within the ellipse and |
2336 | | // compute distances between them and the first point. |
2337 | 0 | { |
2338 | 0 | const int ji = papsPoints[j]->i; |
2339 | 0 | double dfRX2 = padfX[ji] - dfXPoint; |
2340 | 0 | double dfRY2 = padfY[ji] - dfYPoint; |
2341 | |
|
2342 | 0 | if (dfRadius2Square * dfRX2 * dfRX2 + |
2343 | 0 | dfRadius1Square * dfRY2 * dfRY2 <= |
2344 | 0 | dfR12Square) |
2345 | 0 | { |
2346 | 0 | const double dfRX = padfX[ji] - padfX[i]; |
2347 | 0 | const double dfRY = padfY[ji] - padfY[i]; |
2348 | |
|
2349 | 0 | dfAccumulator += sqrt(dfRX * dfRX + dfRY * dfRY); |
2350 | 0 | n++; |
2351 | 0 | } |
2352 | 0 | } |
2353 | 0 | } |
2354 | 0 | } |
2355 | 0 | } |
2356 | 0 | CPLFree(papsPoints); |
2357 | 0 | } |
2358 | 0 | else |
2359 | 0 | { |
2360 | 0 | GUInt32 i = 0; |
2361 | 0 | while (i < nPoints - 1) |
2362 | 0 | { |
2363 | 0 | double dfRX1 = padfX[i] - dfXPoint; |
2364 | 0 | double dfRY1 = padfY[i] - dfYPoint; |
2365 | |
|
2366 | 0 | if (bRotated) |
2367 | 0 | { |
2368 | 0 | const double dfRXRotated = dfRX1 * dfCoeff1 + dfRY1 * dfCoeff2; |
2369 | 0 | const double dfRYRotated = dfRY1 * dfCoeff1 - dfRX1 * dfCoeff2; |
2370 | |
|
2371 | 0 | dfRX1 = dfRXRotated; |
2372 | 0 | dfRY1 = dfRYRotated; |
2373 | 0 | } |
2374 | | |
2375 | | // Is this point located inside the search ellipse? |
2376 | 0 | if (dfRadius2Square * dfRX1 * dfRX1 + |
2377 | 0 | dfRadius1Square * dfRY1 * dfRY1 <= |
2378 | 0 | dfR12Square) |
2379 | 0 | { |
2380 | | // Search all the remaining points within the ellipse and |
2381 | | // compute distances between them and the first point. |
2382 | 0 | for (GUInt32 j = i + 1; j < nPoints; j++) |
2383 | 0 | { |
2384 | 0 | double dfRX2 = padfX[j] - dfXPoint; |
2385 | 0 | double dfRY2 = padfY[j] - dfYPoint; |
2386 | |
|
2387 | 0 | if (bRotated) |
2388 | 0 | { |
2389 | 0 | const double dfRXRotated = |
2390 | 0 | dfRX2 * dfCoeff1 + dfRY2 * dfCoeff2; |
2391 | 0 | const double dfRYRotated = |
2392 | 0 | dfRY2 * dfCoeff1 - dfRX2 * dfCoeff2; |
2393 | |
|
2394 | 0 | dfRX2 = dfRXRotated; |
2395 | 0 | dfRY2 = dfRYRotated; |
2396 | 0 | } |
2397 | |
|
2398 | 0 | if (dfRadius2Square * dfRX2 * dfRX2 + |
2399 | 0 | dfRadius1Square * dfRY2 * dfRY2 <= |
2400 | 0 | dfR12Square) |
2401 | 0 | { |
2402 | 0 | const double dfRX = padfX[j] - padfX[i]; |
2403 | 0 | const double dfRY = padfY[j] - padfY[i]; |
2404 | |
|
2405 | 0 | dfAccumulator += sqrt(dfRX * dfRX + dfRY * dfRY); |
2406 | 0 | n++; |
2407 | 0 | } |
2408 | 0 | } |
2409 | 0 | } |
2410 | |
|
2411 | 0 | i++; |
2412 | 0 | } |
2413 | 0 | } |
2414 | | |
2415 | | // Search for the first point within the search ellipse. |
2416 | 0 | if (n < poOptions->nMinPoints || n == 0) |
2417 | 0 | { |
2418 | 0 | *pdfValue = poOptions->dfNoDataValue; |
2419 | 0 | } |
2420 | 0 | else |
2421 | 0 | { |
2422 | 0 | *pdfValue = dfAccumulator / n; |
2423 | 0 | } |
2424 | |
|
2425 | 0 | return CE_None; |
2426 | 0 | } |
2427 | | |
2428 | | /************************************************************************/ |
2429 | | /* GDALGridLinear() */ |
2430 | | /************************************************************************/ |
2431 | | |
2432 | | /** |
2433 | | * Linear interpolation |
2434 | | * |
2435 | | * The Linear method performs linear interpolation by finding in which triangle |
2436 | | * of a Delaunay triangulation the point is, and by doing interpolation from |
2437 | | * its barycentric coordinates within the triangle. |
2438 | | * If the point is not in any triangle, depending on the radius, the |
2439 | | * algorithm will use the value of the nearest point (radius != 0), |
2440 | | * or the nodata value (radius == 0) |
2441 | | * |
2442 | | * @param poOptionsIn Algorithm parameters. This should point to |
2443 | | * GDALGridLinearOptions object. |
2444 | | * @param nPoints Number of elements in input arrays. |
2445 | | * @param padfX Input array of X coordinates. |
2446 | | * @param padfY Input array of Y coordinates. |
2447 | | * @param padfZ Input array of Z values. |
2448 | | * @param dfXPoint X coordinate of the point to compute. |
2449 | | * @param dfYPoint Y coordinate of the point to compute. |
2450 | | * @param pdfValue Pointer to variable where the computed grid node value |
2451 | | * will be returned. |
2452 | | * @param hExtraParams extra parameters |
2453 | | * |
2454 | | * @return CE_None on success or CE_Failure if something goes wrong. |
2455 | | * |
2456 | | * @since GDAL 2.1 |
2457 | | */ |
2458 | | |
2459 | | CPLErr GDALGridLinear(const void *poOptionsIn, GUInt32 nPoints, |
2460 | | const double *padfX, const double *padfY, |
2461 | | const double *padfZ, double dfXPoint, double dfYPoint, |
2462 | | double *pdfValue, void *hExtraParams) |
2463 | 0 | { |
2464 | 0 | GDALGridExtraParameters *psExtraParams = |
2465 | 0 | static_cast<GDALGridExtraParameters *>(hExtraParams); |
2466 | 0 | GDALTriangulation *psTriangulation = psExtraParams->psTriangulation; |
2467 | |
|
2468 | 0 | int nOutputFacetIdx = -1; |
2469 | 0 | const bool bRet = CPL_TO_BOOL(GDALTriangulationFindFacetDirected( |
2470 | 0 | psTriangulation, psExtraParams->nInitialFacetIdx, dfXPoint, dfYPoint, |
2471 | 0 | &nOutputFacetIdx)); |
2472 | |
|
2473 | 0 | if (bRet) |
2474 | 0 | { |
2475 | 0 | CPLAssert(nOutputFacetIdx >= 0); |
2476 | | // Reuse output facet idx as next initial index since we proceed line by |
2477 | | // line. |
2478 | 0 | psExtraParams->nInitialFacetIdx = nOutputFacetIdx; |
2479 | |
|
2480 | 0 | double lambda1 = 0.0; |
2481 | 0 | double lambda2 = 0.0; |
2482 | 0 | double lambda3 = 0.0; |
2483 | 0 | GDALTriangulationComputeBarycentricCoordinates( |
2484 | 0 | psTriangulation, nOutputFacetIdx, dfXPoint, dfYPoint, &lambda1, |
2485 | 0 | &lambda2, &lambda3); |
2486 | 0 | const int i1 = |
2487 | 0 | psTriangulation->pasFacets[nOutputFacetIdx].anVertexIdx[0]; |
2488 | 0 | const int i2 = |
2489 | 0 | psTriangulation->pasFacets[nOutputFacetIdx].anVertexIdx[1]; |
2490 | 0 | const int i3 = |
2491 | 0 | psTriangulation->pasFacets[nOutputFacetIdx].anVertexIdx[2]; |
2492 | 0 | *pdfValue = |
2493 | 0 | lambda1 * padfZ[i1] + lambda2 * padfZ[i2] + lambda3 * padfZ[i3]; |
2494 | 0 | } |
2495 | 0 | else |
2496 | 0 | { |
2497 | 0 | if (nOutputFacetIdx >= 0) |
2498 | 0 | { |
2499 | | // Also reuse this failed output facet, when valid, as seed for |
2500 | | // next search. |
2501 | 0 | psExtraParams->nInitialFacetIdx = nOutputFacetIdx; |
2502 | 0 | } |
2503 | |
|
2504 | 0 | const GDALGridLinearOptions *const poOptions = |
2505 | 0 | static_cast<const GDALGridLinearOptions *>(poOptionsIn); |
2506 | 0 | const double dfRadius = poOptions->dfRadius; |
2507 | 0 | if (dfRadius == 0.0) |
2508 | 0 | { |
2509 | 0 | *pdfValue = poOptions->dfNoDataValue; |
2510 | 0 | } |
2511 | 0 | else |
2512 | 0 | { |
2513 | 0 | GDALGridNearestNeighborOptions sNeighbourOptions; |
2514 | 0 | sNeighbourOptions.nSizeOfStructure = sizeof(sNeighbourOptions); |
2515 | 0 | sNeighbourOptions.dfRadius1 = |
2516 | 0 | dfRadius < 0.0 || dfRadius >= std::numeric_limits<double>::max() |
2517 | 0 | ? 0.0 |
2518 | 0 | : dfRadius; |
2519 | 0 | sNeighbourOptions.dfRadius2 = |
2520 | 0 | dfRadius < 0.0 || dfRadius >= std::numeric_limits<double>::max() |
2521 | 0 | ? 0.0 |
2522 | 0 | : dfRadius; |
2523 | 0 | sNeighbourOptions.dfAngle = 0.0; |
2524 | 0 | sNeighbourOptions.dfNoDataValue = poOptions->dfNoDataValue; |
2525 | 0 | return GDALGridNearestNeighbor(&sNeighbourOptions, nPoints, padfX, |
2526 | 0 | padfY, padfZ, dfXPoint, dfYPoint, |
2527 | 0 | pdfValue, hExtraParams); |
2528 | 0 | } |
2529 | 0 | } |
2530 | | |
2531 | 0 | return CE_None; |
2532 | 0 | } |
2533 | | |
2534 | | /************************************************************************/ |
2535 | | /* GDALGridJob */ |
2536 | | /************************************************************************/ |
2537 | | |
2538 | | typedef struct _GDALGridJob GDALGridJob; |
2539 | | |
2540 | | struct _GDALGridJob |
2541 | | { |
2542 | | GUInt32 nYStart; |
2543 | | |
2544 | | GByte *pabyData; |
2545 | | GUInt32 nYStep; |
2546 | | GUInt32 nXSize; |
2547 | | GUInt32 nYSize; |
2548 | | double dfXMin; |
2549 | | double dfYMin; |
2550 | | double dfDeltaX; |
2551 | | double dfDeltaY; |
2552 | | GUInt32 nPoints; |
2553 | | const double *padfX; |
2554 | | const double *padfY; |
2555 | | const double *padfZ; |
2556 | | const void *poOptions; |
2557 | | GDALGridFunction pfnGDALGridMethod; |
2558 | | GDALGridExtraParameters *psExtraParameters; |
2559 | | int (*pfnProgress)(GDALGridJob *psJob); |
2560 | | GDALDataType eType; |
2561 | | |
2562 | | int *pnCounter; |
2563 | | int nCounterSingleThreaded; |
2564 | | volatile int *pbStop; |
2565 | | CPLCond *hCond; |
2566 | | CPLMutex *hCondMutex; |
2567 | | |
2568 | | GDALProgressFunc pfnRealProgress; |
2569 | | void *pRealProgressArg; |
2570 | | }; |
2571 | | |
2572 | | /************************************************************************/ |
2573 | | /* GDALGridProgressMultiThread() */ |
2574 | | /************************************************************************/ |
2575 | | |
2576 | | // Return TRUE if the computation must be interrupted. |
2577 | | static int GDALGridProgressMultiThread(GDALGridJob *psJob) |
2578 | 0 | { |
2579 | 0 | CPLAcquireMutex(psJob->hCondMutex, 1.0); |
2580 | 0 | ++(*psJob->pnCounter); |
2581 | 0 | CPLCondSignal(psJob->hCond); |
2582 | 0 | const int bStop = *psJob->pbStop; |
2583 | 0 | CPLReleaseMutex(psJob->hCondMutex); |
2584 | |
|
2585 | 0 | return bStop; |
2586 | 0 | } |
2587 | | |
2588 | | /************************************************************************/ |
2589 | | /* GDALGridProgressMonoThread() */ |
2590 | | /************************************************************************/ |
2591 | | |
2592 | | // Return TRUE if the computation must be interrupted. |
2593 | | static int GDALGridProgressMonoThread(GDALGridJob *psJob) |
2594 | 0 | { |
2595 | 0 | const int nCounter = ++(psJob->nCounterSingleThreaded); |
2596 | 0 | if (!psJob->pfnRealProgress(nCounter / static_cast<double>(psJob->nYSize), |
2597 | 0 | "", psJob->pRealProgressArg)) |
2598 | 0 | { |
2599 | 0 | CPLError(CE_Failure, CPLE_UserInterrupt, "User terminated"); |
2600 | 0 | *psJob->pbStop = TRUE; |
2601 | 0 | return TRUE; |
2602 | 0 | } |
2603 | 0 | return FALSE; |
2604 | 0 | } |
2605 | | |
2606 | | /************************************************************************/ |
2607 | | /* GDALGridJobProcess() */ |
2608 | | /************************************************************************/ |
2609 | | |
2610 | | static void GDALGridJobProcess(void *user_data) |
2611 | 0 | { |
2612 | 0 | GDALGridJob *const psJob = static_cast<GDALGridJob *>(user_data); |
2613 | 0 | int (*pfnProgress)(GDALGridJob * psJob) = psJob->pfnProgress; |
2614 | 0 | const GUInt32 nXSize = psJob->nXSize; |
2615 | | |
2616 | | /* -------------------------------------------------------------------- */ |
2617 | | /* Allocate a buffer of scanline size, fill it with gridded values */ |
2618 | | /* and use GDALCopyWords() to copy values into output data array with */ |
2619 | | /* appropriate data type conversion. */ |
2620 | | /* -------------------------------------------------------------------- */ |
2621 | 0 | double *padfValues = |
2622 | 0 | static_cast<double *>(VSI_MALLOC2_VERBOSE(sizeof(double), nXSize)); |
2623 | 0 | if (padfValues == nullptr) |
2624 | 0 | { |
2625 | 0 | *(psJob->pbStop) = TRUE; |
2626 | 0 | if (pfnProgress != nullptr) |
2627 | 0 | pfnProgress(psJob); // To notify the main thread. |
2628 | 0 | return; |
2629 | 0 | } |
2630 | | |
2631 | 0 | const GUInt32 nYStart = psJob->nYStart; |
2632 | 0 | const GUInt32 nYStep = psJob->nYStep; |
2633 | 0 | GByte *pabyData = psJob->pabyData; |
2634 | |
|
2635 | 0 | const GUInt32 nYSize = psJob->nYSize; |
2636 | 0 | const double dfXMin = psJob->dfXMin; |
2637 | 0 | const double dfYMin = psJob->dfYMin; |
2638 | 0 | const double dfDeltaX = psJob->dfDeltaX; |
2639 | 0 | const double dfDeltaY = psJob->dfDeltaY; |
2640 | 0 | const GUInt32 nPoints = psJob->nPoints; |
2641 | 0 | const double *padfX = psJob->padfX; |
2642 | 0 | const double *padfY = psJob->padfY; |
2643 | 0 | const double *padfZ = psJob->padfZ; |
2644 | 0 | const void *poOptions = psJob->poOptions; |
2645 | 0 | GDALGridFunction pfnGDALGridMethod = psJob->pfnGDALGridMethod; |
2646 | | // Have a local copy of sExtraParameters since we want to modify |
2647 | | // nInitialFacetIdx. |
2648 | 0 | GDALGridExtraParameters sExtraParameters = *psJob->psExtraParameters; |
2649 | 0 | const GDALDataType eType = psJob->eType; |
2650 | |
|
2651 | 0 | const int nDataTypeSize = GDALGetDataTypeSizeBytes(eType); |
2652 | 0 | const int nLineSpace = nXSize * nDataTypeSize; |
2653 | |
|
2654 | 0 | for (GUInt32 nYPoint = nYStart; nYPoint < nYSize; nYPoint += nYStep) |
2655 | 0 | { |
2656 | 0 | const double dfYPoint = dfYMin + (nYPoint + 0.5) * dfDeltaY; |
2657 | |
|
2658 | 0 | for (GUInt32 nXPoint = 0; nXPoint < nXSize; nXPoint++) |
2659 | 0 | { |
2660 | 0 | const double dfXPoint = dfXMin + (nXPoint + 0.5) * dfDeltaX; |
2661 | |
|
2662 | 0 | if ((*pfnGDALGridMethod)(poOptions, nPoints, padfX, padfY, padfZ, |
2663 | 0 | dfXPoint, dfYPoint, padfValues + nXPoint, |
2664 | 0 | &sExtraParameters) != CE_None) |
2665 | 0 | { |
2666 | 0 | CPLError(CE_Failure, CPLE_AppDefined, |
2667 | 0 | "Gridding failed at X position %lu, Y position %lu", |
2668 | 0 | static_cast<long unsigned int>(nXPoint), |
2669 | 0 | static_cast<long unsigned int>(nYPoint)); |
2670 | 0 | *psJob->pbStop = TRUE; |
2671 | 0 | if (pfnProgress != nullptr) |
2672 | 0 | pfnProgress(psJob); // To notify the main thread. |
2673 | 0 | break; |
2674 | 0 | } |
2675 | 0 | } |
2676 | |
|
2677 | 0 | GDALCopyWords(padfValues, GDT_Float64, sizeof(double), |
2678 | 0 | pabyData + nYPoint * nLineSpace, eType, nDataTypeSize, |
2679 | 0 | nXSize); |
2680 | |
|
2681 | 0 | if (*psJob->pbStop || (pfnProgress != nullptr && pfnProgress(psJob))) |
2682 | 0 | break; |
2683 | 0 | } |
2684 | |
|
2685 | 0 | CPLFree(padfValues); |
2686 | 0 | } |
2687 | | |
2688 | | /************************************************************************/ |
2689 | | /* GDALGridContextCreate() */ |
2690 | | /************************************************************************/ |
2691 | | |
2692 | | struct GDALGridContext |
2693 | | { |
2694 | | GDALGridAlgorithm eAlgorithm; |
2695 | | void *poOptions; |
2696 | | GDALGridFunction pfnGDALGridMethod; |
2697 | | |
2698 | | GUInt32 nPoints; |
2699 | | GDALGridPoint *pasGridPoints; |
2700 | | GDALGridXYArrays sXYArrays; |
2701 | | |
2702 | | GDALGridExtraParameters sExtraParameters; |
2703 | | double *padfX; |
2704 | | double *padfY; |
2705 | | double *padfZ; |
2706 | | bool bFreePadfXYZArrays; |
2707 | | |
2708 | | CPLWorkerThreadPool *poWorkerThreadPool; |
2709 | | }; |
2710 | | |
2711 | | static void GDALGridContextCreateQuadTree(GDALGridContext *psContext); |
2712 | | |
2713 | | /** |
2714 | | * Creates a context to do regular gridding from the scattered data. |
2715 | | * |
2716 | | * This function takes the arrays of X and Y coordinates and corresponding Z |
2717 | | * values as input to prepare computation of regular grid (or call it a raster) |
2718 | | * from these scattered data. |
2719 | | * |
2720 | | * On Intel/AMD i386/x86_64 architectures, some |
2721 | | * gridding methods will be optimized with SSE instructions (provided GDAL |
2722 | | * has been compiled with such support, and it is available at runtime). |
2723 | | * Currently, only 'invdist' algorithm with default parameters has an optimized |
2724 | | * implementation. |
2725 | | * This can provide substantial speed-up, but sometimes at the expense of |
2726 | | * reduced floating point precision. This can be disabled by setting the |
2727 | | * GDAL_USE_SSE configuration option to NO. |
2728 | | * A further optimized version can use the AVX |
2729 | | * instruction set. This can be disabled by setting the GDAL_USE_AVX |
2730 | | * configuration option to NO. |
2731 | | * |
2732 | | * It is possible to set the GDAL_NUM_THREADS |
2733 | | * configuration option to parallelize the processing. The value to set is |
2734 | | * the number of worker threads, or ALL_CPUS to use all the cores/CPUs of the |
2735 | | * computer (default value). |
2736 | | * |
2737 | | * @param eAlgorithm Gridding method. |
2738 | | * @param poOptions Options to control chosen gridding method. |
2739 | | * @param nPoints Number of elements in input arrays. |
2740 | | * @param padfX Input array of X coordinates. |
2741 | | * @param padfY Input array of Y coordinates. |
2742 | | * @param padfZ Input array of Z values. |
2743 | | * @param bCallerWillKeepPointArraysAlive Whether the provided padfX, padfY, |
2744 | | * padfZ arrays will still be "alive" during the calls to |
2745 | | * GDALGridContextProcess(). Setting to TRUE prevent them from being |
2746 | | * duplicated in the context. If unsure, set to FALSE. |
2747 | | * |
2748 | | * @return the context (to be freed with GDALGridContextFree()) or NULL in case |
2749 | | * or error. |
2750 | | * |
2751 | | * @since GDAL 2.1 |
2752 | | */ |
2753 | | |
2754 | | GDALGridContext *GDALGridContextCreate(GDALGridAlgorithm eAlgorithm, |
2755 | | const void *poOptions, GUInt32 nPoints, |
2756 | | const double *padfX, const double *padfY, |
2757 | | const double *padfZ, |
2758 | | int bCallerWillKeepPointArraysAlive) |
2759 | 0 | { |
2760 | 0 | CPLAssert(poOptions); |
2761 | 0 | CPLAssert(padfX); |
2762 | 0 | CPLAssert(padfY); |
2763 | 0 | CPLAssert(padfZ); |
2764 | 0 | bool bCreateQuadTree = false; |
2765 | |
|
2766 | 0 | const unsigned int nPointCountThreshold = |
2767 | 0 | atoi(CPLGetConfigOption("GDAL_GRID_POINT_COUNT_THRESHOLD", "100")); |
2768 | | |
2769 | | // Starting address aligned on 32-byte boundary for AVX. |
2770 | 0 | float *pafXAligned = nullptr; |
2771 | 0 | float *pafYAligned = nullptr; |
2772 | 0 | float *pafZAligned = nullptr; |
2773 | |
|
2774 | 0 | void *poOptionsNew = nullptr; |
2775 | |
|
2776 | 0 | GDALGridFunction pfnGDALGridMethod = nullptr; |
2777 | |
|
2778 | 0 | switch (eAlgorithm) |
2779 | 0 | { |
2780 | 0 | case GGA_InverseDistanceToAPower: |
2781 | 0 | { |
2782 | 0 | const auto poOptionsOld = |
2783 | 0 | static_cast<const GDALGridInverseDistanceToAPowerOptions *>( |
2784 | 0 | poOptions); |
2785 | 0 | if (poOptionsOld->nSizeOfStructure != sizeof(*poOptionsOld)) |
2786 | 0 | { |
2787 | 0 | CPLError(CE_Failure, CPLE_AppDefined, |
2788 | 0 | "Wrong value of nSizeOfStructure member"); |
2789 | 0 | return nullptr; |
2790 | 0 | } |
2791 | 0 | poOptionsNew = |
2792 | 0 | CPLMalloc(sizeof(GDALGridInverseDistanceToAPowerOptions)); |
2793 | 0 | memcpy(poOptionsNew, poOptions, |
2794 | 0 | sizeof(GDALGridInverseDistanceToAPowerOptions)); |
2795 | |
|
2796 | 0 | const GDALGridInverseDistanceToAPowerOptions *const poPower = |
2797 | 0 | static_cast<const GDALGridInverseDistanceToAPowerOptions *>( |
2798 | 0 | poOptions); |
2799 | 0 | if (poPower->dfRadius1 == 0.0 && poPower->dfRadius2 == 0.0) |
2800 | 0 | { |
2801 | 0 | const double dfPower = poPower->dfPower; |
2802 | 0 | const double dfSmoothing = poPower->dfSmoothing; |
2803 | |
|
2804 | 0 | pfnGDALGridMethod = GDALGridInverseDistanceToAPowerNoSearch; |
2805 | 0 | if (dfPower == 2.0 && dfSmoothing == 0.0) |
2806 | 0 | { |
2807 | 0 | #ifdef HAVE_AVX_AT_COMPILE_TIME |
2808 | |
|
2809 | 0 | if (CPLTestBool( |
2810 | 0 | CPLGetConfigOption("GDAL_USE_AVX", "YES")) && |
2811 | 0 | CPLHaveRuntimeAVX()) |
2812 | 0 | { |
2813 | 0 | pafXAligned = static_cast<float *>( |
2814 | 0 | VSI_MALLOC_ALIGNED_AUTO_VERBOSE(sizeof(float) * |
2815 | 0 | nPoints)); |
2816 | 0 | pafYAligned = static_cast<float *>( |
2817 | 0 | VSI_MALLOC_ALIGNED_AUTO_VERBOSE(sizeof(float) * |
2818 | 0 | nPoints)); |
2819 | 0 | pafZAligned = static_cast<float *>( |
2820 | 0 | VSI_MALLOC_ALIGNED_AUTO_VERBOSE(sizeof(float) * |
2821 | 0 | nPoints)); |
2822 | 0 | if (pafXAligned != nullptr && pafYAligned != nullptr && |
2823 | 0 | pafZAligned != nullptr) |
2824 | 0 | { |
2825 | 0 | CPLDebug("GDAL_GRID", |
2826 | 0 | "Using AVX optimized version"); |
2827 | 0 | pfnGDALGridMethod = |
2828 | 0 | GDALGridInverseDistanceToAPower2NoSmoothingNoSearchAVX; |
2829 | 0 | for (GUInt32 i = 0; i < nPoints; i++) |
2830 | 0 | { |
2831 | 0 | pafXAligned[i] = static_cast<float>(padfX[i]); |
2832 | 0 | pafYAligned[i] = static_cast<float>(padfY[i]); |
2833 | 0 | pafZAligned[i] = static_cast<float>(padfZ[i]); |
2834 | 0 | } |
2835 | 0 | } |
2836 | 0 | else |
2837 | 0 | { |
2838 | 0 | VSIFree(pafXAligned); |
2839 | 0 | VSIFree(pafYAligned); |
2840 | 0 | VSIFree(pafZAligned); |
2841 | 0 | pafXAligned = nullptr; |
2842 | 0 | pafYAligned = nullptr; |
2843 | 0 | pafZAligned = nullptr; |
2844 | 0 | } |
2845 | 0 | } |
2846 | 0 | #endif |
2847 | |
|
2848 | 0 | #ifdef HAVE_SSE_AT_COMPILE_TIME |
2849 | |
|
2850 | 0 | if (pafXAligned == nullptr && |
2851 | 0 | CPLTestBool(CPLGetConfigOption("GDAL_USE_SSE", "YES")) |
2852 | 0 | #if !defined(USE_NEON_OPTIMIZATIONS) |
2853 | 0 | && CPLHaveRuntimeSSE() |
2854 | 0 | #endif |
2855 | 0 | ) |
2856 | 0 | { |
2857 | 0 | pafXAligned = static_cast<float *>( |
2858 | 0 | VSI_MALLOC_ALIGNED_AUTO_VERBOSE(sizeof(float) * |
2859 | 0 | nPoints)); |
2860 | 0 | pafYAligned = static_cast<float *>( |
2861 | 0 | VSI_MALLOC_ALIGNED_AUTO_VERBOSE(sizeof(float) * |
2862 | 0 | nPoints)); |
2863 | 0 | pafZAligned = static_cast<float *>( |
2864 | 0 | VSI_MALLOC_ALIGNED_AUTO_VERBOSE(sizeof(float) * |
2865 | 0 | nPoints)); |
2866 | 0 | if (pafXAligned != nullptr && pafYAligned != nullptr && |
2867 | 0 | pafZAligned != nullptr) |
2868 | 0 | { |
2869 | 0 | CPLDebug("GDAL_GRID", |
2870 | 0 | "Using SSE optimized version"); |
2871 | 0 | pfnGDALGridMethod = |
2872 | 0 | GDALGridInverseDistanceToAPower2NoSmoothingNoSearchSSE; |
2873 | 0 | for (GUInt32 i = 0; i < nPoints; i++) |
2874 | 0 | { |
2875 | 0 | pafXAligned[i] = static_cast<float>(padfX[i]); |
2876 | 0 | pafYAligned[i] = static_cast<float>(padfY[i]); |
2877 | 0 | pafZAligned[i] = static_cast<float>(padfZ[i]); |
2878 | 0 | } |
2879 | 0 | } |
2880 | 0 | else |
2881 | 0 | { |
2882 | 0 | VSIFree(pafXAligned); |
2883 | 0 | VSIFree(pafYAligned); |
2884 | 0 | VSIFree(pafZAligned); |
2885 | 0 | pafXAligned = nullptr; |
2886 | 0 | pafYAligned = nullptr; |
2887 | 0 | pafZAligned = nullptr; |
2888 | 0 | } |
2889 | 0 | } |
2890 | 0 | #endif // HAVE_SSE_AT_COMPILE_TIME |
2891 | 0 | } |
2892 | 0 | } |
2893 | 0 | else |
2894 | 0 | { |
2895 | 0 | pfnGDALGridMethod = GDALGridInverseDistanceToAPower; |
2896 | 0 | } |
2897 | 0 | break; |
2898 | 0 | } |
2899 | 0 | case GGA_InverseDistanceToAPowerNearestNeighbor: |
2900 | 0 | { |
2901 | 0 | const auto poOptionsOld = static_cast< |
2902 | 0 | const GDALGridInverseDistanceToAPowerNearestNeighborOptions *>( |
2903 | 0 | poOptions); |
2904 | 0 | if (poOptionsOld->nSizeOfStructure != sizeof(*poOptionsOld)) |
2905 | 0 | { |
2906 | 0 | CPLError(CE_Failure, CPLE_AppDefined, |
2907 | 0 | "Wrong value of nSizeOfStructure member"); |
2908 | 0 | return nullptr; |
2909 | 0 | } |
2910 | 0 | poOptionsNew = CPLMalloc( |
2911 | 0 | sizeof(GDALGridInverseDistanceToAPowerNearestNeighborOptions)); |
2912 | 0 | memcpy( |
2913 | 0 | poOptionsNew, poOptions, |
2914 | 0 | sizeof(GDALGridInverseDistanceToAPowerNearestNeighborOptions)); |
2915 | |
|
2916 | 0 | if (poOptionsOld->nMinPointsPerQuadrant != 0 || |
2917 | 0 | poOptionsOld->nMaxPointsPerQuadrant != 0) |
2918 | 0 | { |
2919 | 0 | pfnGDALGridMethod = |
2920 | 0 | GDALGridInverseDistanceToAPowerNearestNeighborPerQuadrant; |
2921 | 0 | } |
2922 | 0 | else |
2923 | 0 | { |
2924 | 0 | pfnGDALGridMethod = |
2925 | 0 | GDALGridInverseDistanceToAPowerNearestNeighbor; |
2926 | 0 | } |
2927 | 0 | bCreateQuadTree = true; |
2928 | 0 | break; |
2929 | 0 | } |
2930 | 0 | case GGA_MovingAverage: |
2931 | 0 | { |
2932 | 0 | const auto poOptionsOld = |
2933 | 0 | static_cast<const GDALGridMovingAverageOptions *>(poOptions); |
2934 | 0 | if (poOptionsOld->nSizeOfStructure != sizeof(*poOptionsOld)) |
2935 | 0 | { |
2936 | 0 | CPLError(CE_Failure, CPLE_AppDefined, |
2937 | 0 | "Wrong value of nSizeOfStructure member"); |
2938 | 0 | return nullptr; |
2939 | 0 | } |
2940 | 0 | poOptionsNew = CPLMalloc(sizeof(GDALGridMovingAverageOptions)); |
2941 | 0 | memcpy(poOptionsNew, poOptions, |
2942 | 0 | sizeof(GDALGridMovingAverageOptions)); |
2943 | |
|
2944 | 0 | if (poOptionsOld->nMinPointsPerQuadrant != 0 || |
2945 | 0 | poOptionsOld->nMaxPointsPerQuadrant != 0) |
2946 | 0 | { |
2947 | 0 | pfnGDALGridMethod = GDALGridMovingAveragePerQuadrant; |
2948 | 0 | bCreateQuadTree = true; |
2949 | 0 | } |
2950 | 0 | else |
2951 | 0 | { |
2952 | 0 | pfnGDALGridMethod = GDALGridMovingAverage; |
2953 | 0 | bCreateQuadTree = (nPoints > nPointCountThreshold && |
2954 | 0 | poOptionsOld->dfAngle == 0.0 && |
2955 | 0 | (poOptionsOld->dfRadius1 > 0.0 || |
2956 | 0 | poOptionsOld->dfRadius2 > 0.0)); |
2957 | 0 | } |
2958 | 0 | break; |
2959 | 0 | } |
2960 | 0 | case GGA_NearestNeighbor: |
2961 | 0 | { |
2962 | 0 | const auto poOptionsOld = |
2963 | 0 | static_cast<const GDALGridNearestNeighborOptions *>(poOptions); |
2964 | 0 | if (poOptionsOld->nSizeOfStructure != sizeof(*poOptionsOld)) |
2965 | 0 | { |
2966 | 0 | CPLError(CE_Failure, CPLE_AppDefined, |
2967 | 0 | "Wrong value of nSizeOfStructure member"); |
2968 | 0 | return nullptr; |
2969 | 0 | } |
2970 | 0 | poOptionsNew = CPLMalloc(sizeof(GDALGridNearestNeighborOptions)); |
2971 | 0 | memcpy(poOptionsNew, poOptions, |
2972 | 0 | sizeof(GDALGridNearestNeighborOptions)); |
2973 | |
|
2974 | 0 | pfnGDALGridMethod = GDALGridNearestNeighbor; |
2975 | 0 | bCreateQuadTree = (nPoints > nPointCountThreshold && |
2976 | 0 | poOptionsOld->dfAngle == 0.0 && |
2977 | 0 | (poOptionsOld->dfRadius1 > 0.0 || |
2978 | 0 | poOptionsOld->dfRadius2 > 0.0)); |
2979 | 0 | break; |
2980 | 0 | } |
2981 | 0 | case GGA_MetricMinimum: |
2982 | 0 | { |
2983 | 0 | const auto poOptionsOld = |
2984 | 0 | static_cast<const GDALGridDataMetricsOptions *>(poOptions); |
2985 | 0 | if (poOptionsOld->nSizeOfStructure != sizeof(*poOptionsOld)) |
2986 | 0 | { |
2987 | 0 | CPLError(CE_Failure, CPLE_AppDefined, |
2988 | 0 | "Wrong value of nSizeOfStructure member"); |
2989 | 0 | return nullptr; |
2990 | 0 | } |
2991 | 0 | poOptionsNew = CPLMalloc(sizeof(GDALGridDataMetricsOptions)); |
2992 | 0 | memcpy(poOptionsNew, poOptions, sizeof(GDALGridDataMetricsOptions)); |
2993 | |
|
2994 | 0 | if (poOptionsOld->nMinPointsPerQuadrant != 0 || |
2995 | 0 | poOptionsOld->nMaxPointsPerQuadrant != 0) |
2996 | 0 | { |
2997 | 0 | pfnGDALGridMethod = GDALGridDataMetricMinimumPerQuadrant; |
2998 | 0 | bCreateQuadTree = true; |
2999 | 0 | } |
3000 | 0 | else |
3001 | 0 | { |
3002 | 0 | pfnGDALGridMethod = GDALGridDataMetricMinimum; |
3003 | 0 | bCreateQuadTree = (nPoints > nPointCountThreshold && |
3004 | 0 | poOptionsOld->dfAngle == 0.0 && |
3005 | 0 | (poOptionsOld->dfRadius1 > 0.0 || |
3006 | 0 | poOptionsOld->dfRadius2 > 0.0)); |
3007 | 0 | } |
3008 | 0 | break; |
3009 | 0 | } |
3010 | 0 | case GGA_MetricMaximum: |
3011 | 0 | { |
3012 | 0 | const auto poOptionsOld = |
3013 | 0 | static_cast<const GDALGridDataMetricsOptions *>(poOptions); |
3014 | 0 | if (poOptionsOld->nSizeOfStructure != sizeof(*poOptionsOld)) |
3015 | 0 | { |
3016 | 0 | CPLError(CE_Failure, CPLE_AppDefined, |
3017 | 0 | "Wrong value of nSizeOfStructure member"); |
3018 | 0 | return nullptr; |
3019 | 0 | } |
3020 | 0 | poOptionsNew = CPLMalloc(sizeof(GDALGridDataMetricsOptions)); |
3021 | 0 | memcpy(poOptionsNew, poOptions, sizeof(GDALGridDataMetricsOptions)); |
3022 | |
|
3023 | 0 | if (poOptionsOld->nMinPointsPerQuadrant != 0 || |
3024 | 0 | poOptionsOld->nMaxPointsPerQuadrant != 0) |
3025 | 0 | { |
3026 | 0 | pfnGDALGridMethod = GDALGridDataMetricMaximumPerQuadrant; |
3027 | 0 | bCreateQuadTree = true; |
3028 | 0 | } |
3029 | 0 | else |
3030 | 0 | { |
3031 | 0 | pfnGDALGridMethod = GDALGridDataMetricMaximum; |
3032 | 0 | bCreateQuadTree = (nPoints > nPointCountThreshold && |
3033 | 0 | poOptionsOld->dfAngle == 0.0 && |
3034 | 0 | (poOptionsOld->dfRadius1 > 0.0 || |
3035 | 0 | poOptionsOld->dfRadius2 > 0.0)); |
3036 | 0 | } |
3037 | |
|
3038 | 0 | break; |
3039 | 0 | } |
3040 | 0 | case GGA_MetricRange: |
3041 | 0 | { |
3042 | 0 | const auto poOptionsOld = |
3043 | 0 | static_cast<const GDALGridDataMetricsOptions *>(poOptions); |
3044 | 0 | if (poOptionsOld->nSizeOfStructure != sizeof(*poOptionsOld)) |
3045 | 0 | { |
3046 | 0 | CPLError(CE_Failure, CPLE_AppDefined, |
3047 | 0 | "Wrong value of nSizeOfStructure member"); |
3048 | 0 | return nullptr; |
3049 | 0 | } |
3050 | 0 | poOptionsNew = CPLMalloc(sizeof(GDALGridDataMetricsOptions)); |
3051 | 0 | memcpy(poOptionsNew, poOptions, sizeof(GDALGridDataMetricsOptions)); |
3052 | |
|
3053 | 0 | if (poOptionsOld->nMinPointsPerQuadrant != 0 || |
3054 | 0 | poOptionsOld->nMaxPointsPerQuadrant != 0) |
3055 | 0 | { |
3056 | 0 | pfnGDALGridMethod = GDALGridDataMetricRangePerQuadrant; |
3057 | 0 | bCreateQuadTree = true; |
3058 | 0 | } |
3059 | 0 | else |
3060 | 0 | { |
3061 | 0 | pfnGDALGridMethod = GDALGridDataMetricRange; |
3062 | 0 | bCreateQuadTree = (nPoints > nPointCountThreshold && |
3063 | 0 | poOptionsOld->dfAngle == 0.0 && |
3064 | 0 | (poOptionsOld->dfRadius1 > 0.0 || |
3065 | 0 | poOptionsOld->dfRadius2 > 0.0)); |
3066 | 0 | } |
3067 | |
|
3068 | 0 | break; |
3069 | 0 | } |
3070 | 0 | case GGA_MetricCount: |
3071 | 0 | { |
3072 | 0 | const auto poOptionsOld = |
3073 | 0 | static_cast<const GDALGridDataMetricsOptions *>(poOptions); |
3074 | 0 | if (poOptionsOld->nSizeOfStructure != sizeof(*poOptionsOld)) |
3075 | 0 | { |
3076 | 0 | CPLError(CE_Failure, CPLE_AppDefined, |
3077 | 0 | "Wrong value of nSizeOfStructure member"); |
3078 | 0 | return nullptr; |
3079 | 0 | } |
3080 | 0 | poOptionsNew = CPLMalloc(sizeof(GDALGridDataMetricsOptions)); |
3081 | 0 | memcpy(poOptionsNew, poOptions, sizeof(GDALGridDataMetricsOptions)); |
3082 | |
|
3083 | 0 | if (poOptionsOld->nMinPointsPerQuadrant != 0 || |
3084 | 0 | poOptionsOld->nMaxPointsPerQuadrant != 0) |
3085 | 0 | { |
3086 | 0 | pfnGDALGridMethod = GDALGridDataMetricCountPerQuadrant; |
3087 | 0 | bCreateQuadTree = true; |
3088 | 0 | } |
3089 | 0 | else |
3090 | 0 | { |
3091 | 0 | pfnGDALGridMethod = GDALGridDataMetricCount; |
3092 | 0 | bCreateQuadTree = (nPoints > nPointCountThreshold && |
3093 | 0 | poOptionsOld->dfAngle == 0.0 && |
3094 | 0 | (poOptionsOld->dfRadius1 > 0.0 || |
3095 | 0 | poOptionsOld->dfRadius2 > 0.0)); |
3096 | 0 | } |
3097 | |
|
3098 | 0 | break; |
3099 | 0 | } |
3100 | 0 | case GGA_MetricAverageDistance: |
3101 | 0 | { |
3102 | 0 | const auto poOptionsOld = |
3103 | 0 | static_cast<const GDALGridDataMetricsOptions *>(poOptions); |
3104 | 0 | if (poOptionsOld->nSizeOfStructure != sizeof(*poOptionsOld)) |
3105 | 0 | { |
3106 | 0 | CPLError(CE_Failure, CPLE_AppDefined, |
3107 | 0 | "Wrong value of nSizeOfStructure member"); |
3108 | 0 | return nullptr; |
3109 | 0 | } |
3110 | 0 | poOptionsNew = CPLMalloc(sizeof(GDALGridDataMetricsOptions)); |
3111 | 0 | memcpy(poOptionsNew, poOptions, sizeof(GDALGridDataMetricsOptions)); |
3112 | |
|
3113 | 0 | if (poOptionsOld->nMinPointsPerQuadrant != 0 || |
3114 | 0 | poOptionsOld->nMaxPointsPerQuadrant != 0) |
3115 | 0 | { |
3116 | 0 | pfnGDALGridMethod = |
3117 | 0 | GDALGridDataMetricAverageDistancePerQuadrant; |
3118 | 0 | bCreateQuadTree = true; |
3119 | 0 | } |
3120 | 0 | else |
3121 | 0 | { |
3122 | 0 | pfnGDALGridMethod = GDALGridDataMetricAverageDistance; |
3123 | 0 | bCreateQuadTree = (nPoints > nPointCountThreshold && |
3124 | 0 | poOptionsOld->dfAngle == 0.0 && |
3125 | 0 | (poOptionsOld->dfRadius1 > 0.0 || |
3126 | 0 | poOptionsOld->dfRadius2 > 0.0)); |
3127 | 0 | } |
3128 | |
|
3129 | 0 | break; |
3130 | 0 | } |
3131 | 0 | case GGA_MetricAverageDistancePts: |
3132 | 0 | { |
3133 | 0 | const auto poOptionsOld = |
3134 | 0 | static_cast<const GDALGridDataMetricsOptions *>(poOptions); |
3135 | 0 | if (poOptionsOld->nSizeOfStructure != sizeof(*poOptionsOld)) |
3136 | 0 | { |
3137 | 0 | CPLError(CE_Failure, CPLE_AppDefined, |
3138 | 0 | "Wrong value of nSizeOfStructure member"); |
3139 | 0 | return nullptr; |
3140 | 0 | } |
3141 | 0 | poOptionsNew = CPLMalloc(sizeof(GDALGridDataMetricsOptions)); |
3142 | 0 | memcpy(poOptionsNew, poOptions, sizeof(GDALGridDataMetricsOptions)); |
3143 | |
|
3144 | 0 | pfnGDALGridMethod = GDALGridDataMetricAverageDistancePts; |
3145 | 0 | bCreateQuadTree = (nPoints > nPointCountThreshold && |
3146 | 0 | poOptionsOld->dfAngle == 0.0 && |
3147 | 0 | (poOptionsOld->dfRadius1 > 0.0 || |
3148 | 0 | poOptionsOld->dfRadius2 > 0.0)); |
3149 | |
|
3150 | 0 | break; |
3151 | 0 | } |
3152 | 0 | case GGA_Linear: |
3153 | 0 | { |
3154 | 0 | const auto poOptionsOld = |
3155 | 0 | static_cast<const GDALGridLinearOptions *>(poOptions); |
3156 | 0 | if (poOptionsOld->nSizeOfStructure != sizeof(*poOptionsOld)) |
3157 | 0 | { |
3158 | 0 | CPLError(CE_Failure, CPLE_AppDefined, |
3159 | 0 | "Wrong value of nSizeOfStructure member"); |
3160 | 0 | return nullptr; |
3161 | 0 | } |
3162 | 0 | poOptionsNew = CPLMalloc(sizeof(GDALGridLinearOptions)); |
3163 | 0 | memcpy(poOptionsNew, poOptions, sizeof(GDALGridLinearOptions)); |
3164 | |
|
3165 | 0 | pfnGDALGridMethod = GDALGridLinear; |
3166 | 0 | break; |
3167 | 0 | } |
3168 | 0 | default: |
3169 | 0 | CPLError(CE_Failure, CPLE_IllegalArg, |
3170 | 0 | "GDAL does not support gridding method %d", eAlgorithm); |
3171 | 0 | return nullptr; |
3172 | 0 | } |
3173 | | |
3174 | 0 | if (pafXAligned == nullptr && !bCallerWillKeepPointArraysAlive) |
3175 | 0 | { |
3176 | 0 | double *padfXNew = |
3177 | 0 | static_cast<double *>(VSI_MALLOC2_VERBOSE(nPoints, sizeof(double))); |
3178 | 0 | double *padfYNew = |
3179 | 0 | static_cast<double *>(VSI_MALLOC2_VERBOSE(nPoints, sizeof(double))); |
3180 | 0 | double *padfZNew = |
3181 | 0 | static_cast<double *>(VSI_MALLOC2_VERBOSE(nPoints, sizeof(double))); |
3182 | 0 | if (padfXNew == nullptr || padfYNew == nullptr || padfZNew == nullptr) |
3183 | 0 | { |
3184 | 0 | VSIFree(padfXNew); |
3185 | 0 | VSIFree(padfYNew); |
3186 | 0 | VSIFree(padfZNew); |
3187 | 0 | CPLFree(poOptionsNew); |
3188 | 0 | return nullptr; |
3189 | 0 | } |
3190 | 0 | memcpy(padfXNew, padfX, nPoints * sizeof(double)); |
3191 | 0 | memcpy(padfYNew, padfY, nPoints * sizeof(double)); |
3192 | 0 | memcpy(padfZNew, padfZ, nPoints * sizeof(double)); |
3193 | 0 | padfX = padfXNew; |
3194 | 0 | padfY = padfYNew; |
3195 | 0 | padfZ = padfZNew; |
3196 | 0 | } |
3197 | 0 | GDALGridContext *psContext = |
3198 | 0 | static_cast<GDALGridContext *>(CPLCalloc(1, sizeof(GDALGridContext))); |
3199 | 0 | psContext->eAlgorithm = eAlgorithm; |
3200 | 0 | psContext->poOptions = poOptionsNew; |
3201 | 0 | psContext->pfnGDALGridMethod = pfnGDALGridMethod; |
3202 | 0 | psContext->nPoints = nPoints; |
3203 | 0 | psContext->pasGridPoints = nullptr; |
3204 | 0 | psContext->sXYArrays.padfX = padfX; |
3205 | 0 | psContext->sXYArrays.padfY = padfY; |
3206 | 0 | psContext->sExtraParameters.hQuadTree = nullptr; |
3207 | 0 | psContext->sExtraParameters.dfInitialSearchRadius = 0.0; |
3208 | 0 | psContext->sExtraParameters.pafX = pafXAligned; |
3209 | 0 | psContext->sExtraParameters.pafY = pafYAligned; |
3210 | 0 | psContext->sExtraParameters.pafZ = pafZAligned; |
3211 | 0 | psContext->sExtraParameters.psTriangulation = nullptr; |
3212 | 0 | psContext->sExtraParameters.nInitialFacetIdx = 0; |
3213 | 0 | psContext->padfX = pafXAligned ? nullptr : const_cast<double *>(padfX); |
3214 | 0 | psContext->padfY = pafXAligned ? nullptr : const_cast<double *>(padfY); |
3215 | 0 | psContext->padfZ = pafXAligned ? nullptr : const_cast<double *>(padfZ); |
3216 | 0 | psContext->bFreePadfXYZArrays = |
3217 | 0 | pafXAligned ? false : !bCallerWillKeepPointArraysAlive; |
3218 | | |
3219 | | /* -------------------------------------------------------------------- */ |
3220 | | /* Create quadtree if requested and possible. */ |
3221 | | /* -------------------------------------------------------------------- */ |
3222 | 0 | if (bCreateQuadTree) |
3223 | 0 | { |
3224 | 0 | GDALGridContextCreateQuadTree(psContext); |
3225 | 0 | if (psContext->sExtraParameters.hQuadTree == nullptr && |
3226 | 0 | (eAlgorithm == GGA_InverseDistanceToAPowerNearestNeighbor || |
3227 | 0 | pfnGDALGridMethod == GDALGridMovingAveragePerQuadrant)) |
3228 | 0 | { |
3229 | | // shouldn't happen unless memory allocation failure occurs |
3230 | 0 | GDALGridContextFree(psContext); |
3231 | 0 | return nullptr; |
3232 | 0 | } |
3233 | 0 | } |
3234 | | |
3235 | | /* -------------------------------------------------------------------- */ |
3236 | | /* Pre-compute extra parameters in GDALGridExtraParameters */ |
3237 | | /* -------------------------------------------------------------------- */ |
3238 | 0 | if (eAlgorithm == GGA_InverseDistanceToAPowerNearestNeighbor) |
3239 | 0 | { |
3240 | 0 | const double dfPower = |
3241 | 0 | static_cast< |
3242 | 0 | const GDALGridInverseDistanceToAPowerNearestNeighborOptions *>( |
3243 | 0 | poOptions) |
3244 | 0 | ->dfPower; |
3245 | 0 | psContext->sExtraParameters.dfPowerDiv2PreComp = dfPower / 2; |
3246 | |
|
3247 | 0 | const double dfRadius = |
3248 | 0 | static_cast< |
3249 | 0 | const GDALGridInverseDistanceToAPowerNearestNeighborOptions *>( |
3250 | 0 | poOptions) |
3251 | 0 | ->dfRadius; |
3252 | 0 | psContext->sExtraParameters.dfRadiusPower2PreComp = pow(dfRadius, 2); |
3253 | 0 | } |
3254 | |
|
3255 | 0 | if (eAlgorithm == GGA_Linear) |
3256 | 0 | { |
3257 | 0 | psContext->sExtraParameters.psTriangulation = |
3258 | 0 | GDALTriangulationCreateDelaunay(nPoints, padfX, padfY); |
3259 | 0 | if (psContext->sExtraParameters.psTriangulation == nullptr) |
3260 | 0 | { |
3261 | 0 | GDALGridContextFree(psContext); |
3262 | 0 | return nullptr; |
3263 | 0 | } |
3264 | 0 | GDALTriangulationComputeBarycentricCoefficients( |
3265 | 0 | psContext->sExtraParameters.psTriangulation, padfX, padfY); |
3266 | 0 | } |
3267 | | |
3268 | | /* -------------------------------------------------------------------- */ |
3269 | | /* Start thread pool. */ |
3270 | | /* -------------------------------------------------------------------- */ |
3271 | 0 | const char *pszThreads = CPLGetConfigOption("GDAL_NUM_THREADS", "ALL_CPUS"); |
3272 | 0 | int nThreads = 0; |
3273 | 0 | if (EQUAL(pszThreads, "ALL_CPUS")) |
3274 | 0 | nThreads = CPLGetNumCPUs(); |
3275 | 0 | else |
3276 | 0 | nThreads = atoi(pszThreads); |
3277 | 0 | if (nThreads > 128) |
3278 | 0 | nThreads = 128; |
3279 | 0 | if (nThreads > 1) |
3280 | 0 | { |
3281 | 0 | psContext->poWorkerThreadPool = new CPLWorkerThreadPool(); |
3282 | 0 | if (!psContext->poWorkerThreadPool->Setup(nThreads, nullptr, nullptr)) |
3283 | 0 | { |
3284 | 0 | delete psContext->poWorkerThreadPool; |
3285 | 0 | psContext->poWorkerThreadPool = nullptr; |
3286 | 0 | } |
3287 | 0 | else |
3288 | 0 | { |
3289 | 0 | CPLDebug("GDAL_GRID", "Using %d threads", nThreads); |
3290 | 0 | } |
3291 | 0 | } |
3292 | 0 | else |
3293 | 0 | psContext->poWorkerThreadPool = nullptr; |
3294 | |
|
3295 | 0 | return psContext; |
3296 | 0 | } |
3297 | | |
3298 | | /************************************************************************/ |
3299 | | /* GDALGridContextCreateQuadTree() */ |
3300 | | /************************************************************************/ |
3301 | | |
3302 | | void GDALGridContextCreateQuadTree(GDALGridContext *psContext) |
3303 | 0 | { |
3304 | 0 | const GUInt32 nPoints = psContext->nPoints; |
3305 | 0 | psContext->pasGridPoints = static_cast<GDALGridPoint *>( |
3306 | 0 | VSI_MALLOC2_VERBOSE(nPoints, sizeof(GDALGridPoint))); |
3307 | 0 | if (psContext->pasGridPoints != nullptr) |
3308 | 0 | { |
3309 | 0 | const double *const padfX = psContext->padfX; |
3310 | 0 | const double *const padfY = psContext->padfY; |
3311 | | |
3312 | | // Determine point extents. |
3313 | 0 | CPLRectObj sRect; |
3314 | 0 | sRect.minx = padfX[0]; |
3315 | 0 | sRect.miny = padfY[0]; |
3316 | 0 | sRect.maxx = padfX[0]; |
3317 | 0 | sRect.maxy = padfY[0]; |
3318 | 0 | for (GUInt32 i = 1; i < nPoints; i++) |
3319 | 0 | { |
3320 | 0 | if (padfX[i] < sRect.minx) |
3321 | 0 | sRect.minx = padfX[i]; |
3322 | 0 | if (padfY[i] < sRect.miny) |
3323 | 0 | sRect.miny = padfY[i]; |
3324 | 0 | if (padfX[i] > sRect.maxx) |
3325 | 0 | sRect.maxx = padfX[i]; |
3326 | 0 | if (padfY[i] > sRect.maxy) |
3327 | 0 | sRect.maxy = padfY[i]; |
3328 | 0 | } |
3329 | | |
3330 | | // Initial value for search radius is the typical dimension of a |
3331 | | // "pixel" of the point array (assuming rather uniform distribution). |
3332 | 0 | psContext->sExtraParameters.dfInitialSearchRadius = sqrt( |
3333 | 0 | (sRect.maxx - sRect.minx) * (sRect.maxy - sRect.miny) / nPoints); |
3334 | |
|
3335 | 0 | psContext->sExtraParameters.hQuadTree = |
3336 | 0 | CPLQuadTreeCreate(&sRect, GDALGridGetPointBounds); |
3337 | |
|
3338 | 0 | for (GUInt32 i = 0; i < nPoints; i++) |
3339 | 0 | { |
3340 | 0 | psContext->pasGridPoints[i].psXYArrays = &(psContext->sXYArrays); |
3341 | 0 | psContext->pasGridPoints[i].i = i; |
3342 | 0 | CPLQuadTreeInsert(psContext->sExtraParameters.hQuadTree, |
3343 | 0 | psContext->pasGridPoints + i); |
3344 | 0 | } |
3345 | 0 | } |
3346 | 0 | } |
3347 | | |
3348 | | /************************************************************************/ |
3349 | | /* GDALGridContextFree() */ |
3350 | | /************************************************************************/ |
3351 | | |
3352 | | /** |
3353 | | * Free a context used created by GDALGridContextCreate() |
3354 | | * |
3355 | | * @param psContext the context. |
3356 | | * |
3357 | | * @since GDAL 2.1 |
3358 | | */ |
3359 | | void GDALGridContextFree(GDALGridContext *psContext) |
3360 | 0 | { |
3361 | 0 | if (psContext) |
3362 | 0 | { |
3363 | 0 | CPLFree(psContext->poOptions); |
3364 | 0 | CPLFree(psContext->pasGridPoints); |
3365 | 0 | if (psContext->sExtraParameters.hQuadTree != nullptr) |
3366 | 0 | CPLQuadTreeDestroy(psContext->sExtraParameters.hQuadTree); |
3367 | 0 | if (psContext->bFreePadfXYZArrays) |
3368 | 0 | { |
3369 | 0 | CPLFree(psContext->padfX); |
3370 | 0 | CPLFree(psContext->padfY); |
3371 | 0 | CPLFree(psContext->padfZ); |
3372 | 0 | } |
3373 | 0 | VSIFreeAligned(psContext->sExtraParameters.pafX); |
3374 | 0 | VSIFreeAligned(psContext->sExtraParameters.pafY); |
3375 | 0 | VSIFreeAligned(psContext->sExtraParameters.pafZ); |
3376 | 0 | if (psContext->sExtraParameters.psTriangulation) |
3377 | 0 | GDALTriangulationFree(psContext->sExtraParameters.psTriangulation); |
3378 | 0 | delete psContext->poWorkerThreadPool; |
3379 | 0 | CPLFree(psContext); |
3380 | 0 | } |
3381 | 0 | } |
3382 | | |
3383 | | /************************************************************************/ |
3384 | | /* GDALGridContextProcess() */ |
3385 | | /************************************************************************/ |
3386 | | |
3387 | | /** |
3388 | | * Do the gridding of a window of a raster. |
3389 | | * |
3390 | | * This function takes the gridding context as input to preprare computation |
3391 | | * of regular grid (or call it a raster) from these scattered data. |
3392 | | * You should supply the extent of the output grid and allocate array |
3393 | | * sufficient to hold such a grid. |
3394 | | * |
3395 | | * @param psContext Gridding context. |
3396 | | * @param dfXMin Lowest X border of output grid. |
3397 | | * @param dfXMax Highest X border of output grid. |
3398 | | * @param dfYMin Lowest Y border of output grid. |
3399 | | * @param dfYMax Highest Y border of output grid. |
3400 | | * @param nXSize Number of columns in output grid. |
3401 | | * @param nYSize Number of rows in output grid. |
3402 | | * @param eType Data type of output array. |
3403 | | * @param pData Pointer to array where the computed grid will be stored. |
3404 | | * @param pfnProgress a GDALProgressFunc() compatible callback function for |
3405 | | * reporting progress or NULL. |
3406 | | * @param pProgressArg argument to be passed to pfnProgress. May be NULL. |
3407 | | * |
3408 | | * @return CE_None on success or CE_Failure if something goes wrong. |
3409 | | * |
3410 | | * @since GDAL 2.1 |
3411 | | */ |
3412 | | |
3413 | | CPLErr GDALGridContextProcess(GDALGridContext *psContext, double dfXMin, |
3414 | | double dfXMax, double dfYMin, double dfYMax, |
3415 | | GUInt32 nXSize, GUInt32 nYSize, |
3416 | | GDALDataType eType, void *pData, |
3417 | | GDALProgressFunc pfnProgress, void *pProgressArg) |
3418 | 0 | { |
3419 | 0 | CPLAssert(psContext); |
3420 | 0 | CPLAssert(pData); |
3421 | |
|
3422 | 0 | if (nXSize == 0 || nYSize == 0) |
3423 | 0 | { |
3424 | 0 | CPLError(CE_Failure, CPLE_IllegalArg, |
3425 | 0 | "Output raster dimensions should have non-zero size."); |
3426 | 0 | return CE_Failure; |
3427 | 0 | } |
3428 | | |
3429 | 0 | const double dfDeltaX = (dfXMax - dfXMin) / nXSize; |
3430 | 0 | const double dfDeltaY = (dfYMax - dfYMin) / nYSize; |
3431 | | |
3432 | | // For linear, check if we will need to fallback to nearest neighbour |
3433 | | // by sampling along the edges. If all points on edges are within |
3434 | | // triangles, then interior points will also be. |
3435 | 0 | if (psContext->eAlgorithm == GGA_Linear && |
3436 | 0 | psContext->sExtraParameters.hQuadTree == nullptr) |
3437 | 0 | { |
3438 | 0 | bool bNeedNearest = false; |
3439 | 0 | int nStartLeft = 0; |
3440 | 0 | int nStartRight = 0; |
3441 | 0 | const double dfXPointMin = dfXMin + (0 + 0.5) * dfDeltaX; |
3442 | 0 | const double dfXPointMax = dfXMin + (nXSize - 1 + 0.5) * dfDeltaX; |
3443 | 0 | for (GUInt32 nYPoint = 0; !bNeedNearest && nYPoint < nYSize; nYPoint++) |
3444 | 0 | { |
3445 | 0 | const double dfYPoint = dfYMin + (nYPoint + 0.5) * dfDeltaY; |
3446 | |
|
3447 | 0 | if (!GDALTriangulationFindFacetDirected( |
3448 | 0 | psContext->sExtraParameters.psTriangulation, nStartLeft, |
3449 | 0 | dfXPointMin, dfYPoint, &nStartLeft)) |
3450 | 0 | { |
3451 | 0 | bNeedNearest = true; |
3452 | 0 | } |
3453 | 0 | if (!GDALTriangulationFindFacetDirected( |
3454 | 0 | psContext->sExtraParameters.psTriangulation, nStartRight, |
3455 | 0 | dfXPointMax, dfYPoint, &nStartRight)) |
3456 | 0 | { |
3457 | 0 | bNeedNearest = true; |
3458 | 0 | } |
3459 | 0 | } |
3460 | 0 | int nStartTop = 0; |
3461 | 0 | int nStartBottom = 0; |
3462 | 0 | const double dfYPointMin = dfYMin + (0 + 0.5) * dfDeltaY; |
3463 | 0 | const double dfYPointMax = dfYMin + (nYSize - 1 + 0.5) * dfDeltaY; |
3464 | 0 | for (GUInt32 nXPoint = 1; !bNeedNearest && nXPoint + 1 < nXSize; |
3465 | 0 | nXPoint++) |
3466 | 0 | { |
3467 | 0 | const double dfXPoint = dfXMin + (nXPoint + 0.5) * dfDeltaX; |
3468 | |
|
3469 | 0 | if (!GDALTriangulationFindFacetDirected( |
3470 | 0 | psContext->sExtraParameters.psTriangulation, nStartTop, |
3471 | 0 | dfXPoint, dfYPointMin, &nStartTop)) |
3472 | 0 | { |
3473 | 0 | bNeedNearest = true; |
3474 | 0 | } |
3475 | 0 | if (!GDALTriangulationFindFacetDirected( |
3476 | 0 | psContext->sExtraParameters.psTriangulation, nStartBottom, |
3477 | 0 | dfXPoint, dfYPointMax, &nStartBottom)) |
3478 | 0 | { |
3479 | 0 | bNeedNearest = true; |
3480 | 0 | } |
3481 | 0 | } |
3482 | 0 | if (bNeedNearest) |
3483 | 0 | { |
3484 | 0 | CPLDebug("GDAL_GRID", "Will need nearest neighbour"); |
3485 | 0 | GDALGridContextCreateQuadTree(psContext); |
3486 | 0 | } |
3487 | 0 | } |
3488 | |
|
3489 | 0 | int nCounter = 0; |
3490 | 0 | volatile int bStop = FALSE; |
3491 | 0 | GDALGridJob sJob; |
3492 | 0 | sJob.nYStart = 0; |
3493 | 0 | sJob.pabyData = static_cast<GByte *>(pData); |
3494 | 0 | sJob.nYStep = 1; |
3495 | 0 | sJob.nXSize = nXSize; |
3496 | 0 | sJob.nYSize = nYSize; |
3497 | 0 | sJob.dfXMin = dfXMin; |
3498 | 0 | sJob.dfYMin = dfYMin; |
3499 | 0 | sJob.dfDeltaX = dfDeltaX; |
3500 | 0 | sJob.dfDeltaY = dfDeltaY; |
3501 | 0 | sJob.nPoints = psContext->nPoints; |
3502 | 0 | sJob.padfX = psContext->padfX; |
3503 | 0 | sJob.padfY = psContext->padfY; |
3504 | 0 | sJob.padfZ = psContext->padfZ; |
3505 | 0 | sJob.poOptions = psContext->poOptions; |
3506 | 0 | sJob.pfnGDALGridMethod = psContext->pfnGDALGridMethod; |
3507 | 0 | sJob.psExtraParameters = &psContext->sExtraParameters; |
3508 | 0 | sJob.pfnProgress = nullptr; |
3509 | 0 | sJob.eType = eType; |
3510 | 0 | sJob.pfnRealProgress = pfnProgress; |
3511 | 0 | sJob.pRealProgressArg = pProgressArg; |
3512 | 0 | sJob.nCounterSingleThreaded = 0; |
3513 | 0 | sJob.pnCounter = &nCounter; |
3514 | 0 | sJob.pbStop = &bStop; |
3515 | 0 | sJob.hCond = nullptr; |
3516 | 0 | sJob.hCondMutex = nullptr; |
3517 | |
|
3518 | 0 | if (psContext->poWorkerThreadPool == nullptr) |
3519 | 0 | { |
3520 | 0 | if (sJob.pfnRealProgress != nullptr && |
3521 | 0 | sJob.pfnRealProgress != GDALDummyProgress) |
3522 | 0 | { |
3523 | 0 | sJob.pfnProgress = GDALGridProgressMonoThread; |
3524 | 0 | } |
3525 | |
|
3526 | 0 | GDALGridJobProcess(&sJob); |
3527 | 0 | } |
3528 | 0 | else |
3529 | 0 | { |
3530 | 0 | int nThreads = psContext->poWorkerThreadPool->GetThreadCount(); |
3531 | 0 | GDALGridJob *pasJobs = static_cast<GDALGridJob *>( |
3532 | 0 | CPLMalloc(sizeof(GDALGridJob) * nThreads)); |
3533 | |
|
3534 | 0 | sJob.nYStep = nThreads; |
3535 | 0 | sJob.hCondMutex = CPLCreateMutex(); /* and implicitly take the mutex */ |
3536 | 0 | sJob.hCond = CPLCreateCond(); |
3537 | 0 | sJob.pfnProgress = GDALGridProgressMultiThread; |
3538 | | |
3539 | | /* -------------------------------------------------------------------- |
3540 | | */ |
3541 | | /* Start threads. */ |
3542 | | /* -------------------------------------------------------------------- |
3543 | | */ |
3544 | 0 | for (int i = 0; i < nThreads && !bStop; i++) |
3545 | 0 | { |
3546 | 0 | memcpy(&pasJobs[i], &sJob, sizeof(GDALGridJob)); |
3547 | 0 | pasJobs[i].nYStart = i; |
3548 | 0 | psContext->poWorkerThreadPool->SubmitJob(GDALGridJobProcess, |
3549 | 0 | &pasJobs[i]); |
3550 | 0 | } |
3551 | | |
3552 | | /* -------------------------------------------------------------------- |
3553 | | */ |
3554 | | /* Report progress. */ |
3555 | | /* -------------------------------------------------------------------- |
3556 | | */ |
3557 | 0 | while (*(sJob.pnCounter) < static_cast<int>(nYSize) && !bStop) |
3558 | 0 | { |
3559 | 0 | CPLCondWait(sJob.hCond, sJob.hCondMutex); |
3560 | |
|
3561 | 0 | int nLocalCounter = *(sJob.pnCounter); |
3562 | 0 | CPLReleaseMutex(sJob.hCondMutex); |
3563 | |
|
3564 | 0 | if (pfnProgress != nullptr && |
3565 | 0 | !pfnProgress(nLocalCounter / static_cast<double>(nYSize), "", |
3566 | 0 | pProgressArg)) |
3567 | 0 | { |
3568 | 0 | CPLError(CE_Failure, CPLE_UserInterrupt, "User terminated"); |
3569 | 0 | bStop = TRUE; |
3570 | 0 | } |
3571 | |
|
3572 | 0 | CPLAcquireMutex(sJob.hCondMutex, 1.0); |
3573 | 0 | } |
3574 | | |
3575 | | // Release mutex before joining threads, otherwise they will dead-lock |
3576 | | // forever in GDALGridProgressMultiThread(). |
3577 | 0 | CPLReleaseMutex(sJob.hCondMutex); |
3578 | | |
3579 | | /* -------------------------------------------------------------------- |
3580 | | */ |
3581 | | /* Wait for all threads to complete and finish. */ |
3582 | | /* -------------------------------------------------------------------- |
3583 | | */ |
3584 | 0 | psContext->poWorkerThreadPool->WaitCompletion(); |
3585 | |
|
3586 | 0 | CPLFree(pasJobs); |
3587 | 0 | CPLDestroyCond(sJob.hCond); |
3588 | 0 | CPLDestroyMutex(sJob.hCondMutex); |
3589 | 0 | } |
3590 | |
|
3591 | 0 | return bStop ? CE_Failure : CE_None; |
3592 | 0 | } |
3593 | | |
3594 | | /************************************************************************/ |
3595 | | /* GDALGridCreate() */ |
3596 | | /************************************************************************/ |
3597 | | |
3598 | | /** |
3599 | | * Create regular grid from the scattered data. |
3600 | | * |
3601 | | * This function takes the arrays of X and Y coordinates and corresponding Z |
3602 | | * values as input and computes regular grid (or call it a raster) from these |
3603 | | * scattered data. You should supply geometry and extent of the output grid |
3604 | | * and allocate array sufficient to hold such a grid. |
3605 | | * |
3606 | | * Starting with GDAL 1.10, it is possible to set the GDAL_NUM_THREADS |
3607 | | * configuration option to parallelize the processing. The value to set is |
3608 | | * the number of worker threads, or ALL_CPUS to use all the cores/CPUs of the |
3609 | | * computer (default value). |
3610 | | * |
3611 | | * Starting with GDAL 1.10, on Intel/AMD i386/x86_64 architectures, some |
3612 | | * gridding methods will be optimized with SSE instructions (provided GDAL |
3613 | | * has been compiled with such support, and it is available at runtime). |
3614 | | * Currently, only 'invdist' algorithm with default parameters has an optimized |
3615 | | * implementation. |
3616 | | * This can provide substantial speed-up, but sometimes at the expense of |
3617 | | * reduced floating point precision. This can be disabled by setting the |
3618 | | * GDAL_USE_SSE configuration option to NO. |
3619 | | * Starting with GDAL 1.11, a further optimized version can use the AVX |
3620 | | * instruction set. This can be disabled by setting the GDAL_USE_AVX |
3621 | | * configuration option to NO. |
3622 | | * |
3623 | | * Note: it will be more efficient to use GDALGridContextCreate(), |
3624 | | * GDALGridContextProcess() and GDALGridContextFree() when doing repeated |
3625 | | * gridding operations with the same algorithm, parameters and points, and |
3626 | | * moving the window in the output grid. |
3627 | | * |
3628 | | * @param eAlgorithm Gridding method. |
3629 | | * @param poOptions Options to control chosen gridding method. |
3630 | | * @param nPoints Number of elements in input arrays. |
3631 | | * @param padfX Input array of X coordinates. |
3632 | | * @param padfY Input array of Y coordinates. |
3633 | | * @param padfZ Input array of Z values. |
3634 | | * @param dfXMin Lowest X border of output grid. |
3635 | | * @param dfXMax Highest X border of output grid. |
3636 | | * @param dfYMin Lowest Y border of output grid. |
3637 | | * @param dfYMax Highest Y border of output grid. |
3638 | | * @param nXSize Number of columns in output grid. |
3639 | | * @param nYSize Number of rows in output grid. |
3640 | | * @param eType Data type of output array. |
3641 | | * @param pData Pointer to array where the computed grid will be stored. |
3642 | | * @param pfnProgress a GDALProgressFunc() compatible callback function for |
3643 | | * reporting progress or NULL. |
3644 | | * @param pProgressArg argument to be passed to pfnProgress. May be NULL. |
3645 | | * |
3646 | | * @return CE_None on success or CE_Failure if something goes wrong. |
3647 | | */ |
3648 | | |
3649 | | CPLErr GDALGridCreate(GDALGridAlgorithm eAlgorithm, const void *poOptions, |
3650 | | GUInt32 nPoints, const double *padfX, const double *padfY, |
3651 | | const double *padfZ, double dfXMin, double dfXMax, |
3652 | | double dfYMin, double dfYMax, GUInt32 nXSize, |
3653 | | GUInt32 nYSize, GDALDataType eType, void *pData, |
3654 | | GDALProgressFunc pfnProgress, void *pProgressArg) |
3655 | 0 | { |
3656 | 0 | GDALGridContext *psContext = GDALGridContextCreate( |
3657 | 0 | eAlgorithm, poOptions, nPoints, padfX, padfY, padfZ, TRUE); |
3658 | 0 | CPLErr eErr = CE_Failure; |
3659 | 0 | if (psContext) |
3660 | 0 | { |
3661 | 0 | eErr = GDALGridContextProcess(psContext, dfXMin, dfXMax, dfYMin, dfYMax, |
3662 | 0 | nXSize, nYSize, eType, pData, pfnProgress, |
3663 | 0 | pProgressArg); |
3664 | 0 | } |
3665 | |
|
3666 | 0 | GDALGridContextFree(psContext); |
3667 | 0 | return eErr; |
3668 | 0 | } |
3669 | | |
3670 | | /************************************************************************/ |
3671 | | /* GDALGridParseAlgorithmAndOptions() */ |
3672 | | /************************************************************************/ |
3673 | | |
3674 | | /** Translates mnemonic gridding algorithm names into GDALGridAlgorithm code, |
3675 | | * parse control parameters and assign defaults. |
3676 | | */ |
3677 | | CPLErr GDALGridParseAlgorithmAndOptions(const char *pszAlgorithm, |
3678 | | GDALGridAlgorithm *peAlgorithm, |
3679 | | void **ppOptions) |
3680 | 0 | { |
3681 | 0 | CPLAssert(pszAlgorithm); |
3682 | 0 | CPLAssert(peAlgorithm); |
3683 | 0 | CPLAssert(ppOptions); |
3684 | |
|
3685 | 0 | *ppOptions = nullptr; |
3686 | |
|
3687 | 0 | char **papszParams = CSLTokenizeString2(pszAlgorithm, ":", FALSE); |
3688 | |
|
3689 | 0 | if (CSLCount(papszParams) < 1) |
3690 | 0 | { |
3691 | 0 | CSLDestroy(papszParams); |
3692 | 0 | return CE_Failure; |
3693 | 0 | } |
3694 | | |
3695 | 0 | if (EQUAL(papszParams[0], szAlgNameInvDist)) |
3696 | 0 | { |
3697 | 0 | if (CSLFetchNameValue(papszParams, "min_points_per_quadrant") || |
3698 | 0 | CSLFetchNameValue(papszParams, "max_points_per_quadrant")) |
3699 | 0 | { |
3700 | | // Remap invdist to invdistnn if per quadrant is required |
3701 | 0 | if (CSLFetchNameValue(papszParams, "radius") == nullptr) |
3702 | 0 | { |
3703 | 0 | const double dfRadius1 = |
3704 | 0 | CPLAtofM(CSLFetchNameValueDef(papszParams, "radius1", "1")); |
3705 | 0 | const double dfRadius2 = |
3706 | 0 | CPLAtofM(CSLFetchNameValueDef(papszParams, "radius2", "1")); |
3707 | 0 | if (dfRadius1 != dfRadius2) |
3708 | 0 | { |
3709 | 0 | CPLError(CE_Failure, CPLE_NotSupported, |
3710 | 0 | "radius1 != radius2 not supported when " |
3711 | 0 | "min_points_per_quadrant and/or " |
3712 | 0 | "max_points_per_quadrant is specified"); |
3713 | 0 | CSLDestroy(papszParams); |
3714 | 0 | return CE_Failure; |
3715 | 0 | } |
3716 | 0 | } |
3717 | | |
3718 | 0 | if (CPLAtofM(CSLFetchNameValueDef(papszParams, "angle", "0")) != 0) |
3719 | 0 | { |
3720 | 0 | CPLError(CE_Failure, CPLE_NotSupported, |
3721 | 0 | "angle != 0 not supported when " |
3722 | 0 | "min_points_per_quadrant and/or " |
3723 | 0 | "max_points_per_quadrant is specified"); |
3724 | 0 | CSLDestroy(papszParams); |
3725 | 0 | return CE_Failure; |
3726 | 0 | } |
3727 | | |
3728 | 0 | char **papszNewParams = CSLAddString(nullptr, "invdistnn"); |
3729 | 0 | if (CSLFetchNameValue(papszParams, "radius") == nullptr) |
3730 | 0 | { |
3731 | 0 | papszNewParams = CSLSetNameValue( |
3732 | 0 | papszNewParams, "radius", |
3733 | 0 | CSLFetchNameValueDef(papszParams, "radius1", "1")); |
3734 | 0 | } |
3735 | 0 | for (const char *pszItem : |
3736 | 0 | {"radius", "power", "smoothing", "max_points", "min_points", |
3737 | 0 | "nodata", "min_points_per_quadrant", |
3738 | 0 | "max_points_per_quadrant"}) |
3739 | 0 | { |
3740 | 0 | const char *pszValue = CSLFetchNameValue(papszParams, pszItem); |
3741 | 0 | if (pszValue) |
3742 | 0 | papszNewParams = |
3743 | 0 | CSLSetNameValue(papszNewParams, pszItem, pszValue); |
3744 | 0 | } |
3745 | 0 | CSLDestroy(papszParams); |
3746 | 0 | papszParams = papszNewParams; |
3747 | |
|
3748 | 0 | *peAlgorithm = GGA_InverseDistanceToAPowerNearestNeighbor; |
3749 | 0 | } |
3750 | 0 | else |
3751 | 0 | { |
3752 | 0 | *peAlgorithm = GGA_InverseDistanceToAPower; |
3753 | 0 | } |
3754 | 0 | } |
3755 | 0 | else if (EQUAL(papszParams[0], szAlgNameInvDistNearestNeighbor)) |
3756 | 0 | { |
3757 | 0 | *peAlgorithm = GGA_InverseDistanceToAPowerNearestNeighbor; |
3758 | 0 | } |
3759 | 0 | else if (EQUAL(papszParams[0], szAlgNameAverage)) |
3760 | 0 | { |
3761 | 0 | *peAlgorithm = GGA_MovingAverage; |
3762 | 0 | } |
3763 | 0 | else if (EQUAL(papszParams[0], szAlgNameNearest)) |
3764 | 0 | { |
3765 | 0 | *peAlgorithm = GGA_NearestNeighbor; |
3766 | 0 | } |
3767 | 0 | else if (EQUAL(papszParams[0], szAlgNameMinimum)) |
3768 | 0 | { |
3769 | 0 | *peAlgorithm = GGA_MetricMinimum; |
3770 | 0 | } |
3771 | 0 | else if (EQUAL(papszParams[0], szAlgNameMaximum)) |
3772 | 0 | { |
3773 | 0 | *peAlgorithm = GGA_MetricMaximum; |
3774 | 0 | } |
3775 | 0 | else if (EQUAL(papszParams[0], szAlgNameRange)) |
3776 | 0 | { |
3777 | 0 | *peAlgorithm = GGA_MetricRange; |
3778 | 0 | } |
3779 | 0 | else if (EQUAL(papszParams[0], szAlgNameCount)) |
3780 | 0 | { |
3781 | 0 | *peAlgorithm = GGA_MetricCount; |
3782 | 0 | } |
3783 | 0 | else if (EQUAL(papszParams[0], szAlgNameAverageDistance)) |
3784 | 0 | { |
3785 | 0 | *peAlgorithm = GGA_MetricAverageDistance; |
3786 | 0 | } |
3787 | 0 | else if (EQUAL(papszParams[0], szAlgNameAverageDistancePts)) |
3788 | 0 | { |
3789 | 0 | *peAlgorithm = GGA_MetricAverageDistancePts; |
3790 | 0 | } |
3791 | 0 | else if (EQUAL(papszParams[0], szAlgNameLinear)) |
3792 | 0 | { |
3793 | 0 | *peAlgorithm = GGA_Linear; |
3794 | 0 | } |
3795 | 0 | else |
3796 | 0 | { |
3797 | 0 | CPLError(CE_Failure, CPLE_IllegalArg, |
3798 | 0 | "Unsupported gridding method \"%s\"", papszParams[0]); |
3799 | 0 | CSLDestroy(papszParams); |
3800 | 0 | return CE_Failure; |
3801 | 0 | } |
3802 | | |
3803 | | /* -------------------------------------------------------------------- */ |
3804 | | /* Parse algorithm parameters and assign defaults. */ |
3805 | | /* -------------------------------------------------------------------- */ |
3806 | 0 | const char *const *papszKnownOptions = nullptr; |
3807 | |
|
3808 | 0 | switch (*peAlgorithm) |
3809 | 0 | { |
3810 | 0 | case GGA_InverseDistanceToAPower: |
3811 | 0 | default: |
3812 | 0 | { |
3813 | 0 | *ppOptions = |
3814 | 0 | CPLMalloc(sizeof(GDALGridInverseDistanceToAPowerOptions)); |
3815 | |
|
3816 | 0 | GDALGridInverseDistanceToAPowerOptions *const poPowerOpts = |
3817 | 0 | static_cast<GDALGridInverseDistanceToAPowerOptions *>( |
3818 | 0 | *ppOptions); |
3819 | |
|
3820 | 0 | poPowerOpts->nSizeOfStructure = sizeof(*poPowerOpts); |
3821 | |
|
3822 | 0 | const char *pszValue = CSLFetchNameValue(papszParams, "power"); |
3823 | 0 | poPowerOpts->dfPower = pszValue ? CPLAtofM(pszValue) : 2.0; |
3824 | |
|
3825 | 0 | pszValue = CSLFetchNameValue(papszParams, "smoothing"); |
3826 | 0 | poPowerOpts->dfSmoothing = pszValue ? CPLAtofM(pszValue) : 0.0; |
3827 | |
|
3828 | 0 | pszValue = CSLFetchNameValue(papszParams, "radius"); |
3829 | 0 | if (pszValue) |
3830 | 0 | { |
3831 | 0 | poPowerOpts->dfRadius1 = CPLAtofM(pszValue); |
3832 | 0 | poPowerOpts->dfRadius2 = poPowerOpts->dfRadius1; |
3833 | 0 | } |
3834 | 0 | else |
3835 | 0 | { |
3836 | 0 | pszValue = CSLFetchNameValue(papszParams, "radius1"); |
3837 | 0 | poPowerOpts->dfRadius1 = pszValue ? CPLAtofM(pszValue) : 0.0; |
3838 | |
|
3839 | 0 | pszValue = CSLFetchNameValue(papszParams, "radius2"); |
3840 | 0 | poPowerOpts->dfRadius2 = pszValue ? CPLAtofM(pszValue) : 0.0; |
3841 | 0 | } |
3842 | |
|
3843 | 0 | pszValue = CSLFetchNameValue(papszParams, "angle"); |
3844 | 0 | poPowerOpts->dfAngle = pszValue ? CPLAtofM(pszValue) : 0.0; |
3845 | |
|
3846 | 0 | pszValue = CSLFetchNameValue(papszParams, "max_points"); |
3847 | 0 | poPowerOpts->nMaxPoints = |
3848 | 0 | static_cast<GUInt32>(pszValue ? CPLAtofM(pszValue) : 0); |
3849 | |
|
3850 | 0 | pszValue = CSLFetchNameValue(papszParams, "min_points"); |
3851 | 0 | poPowerOpts->nMinPoints = |
3852 | 0 | static_cast<GUInt32>(pszValue ? CPLAtofM(pszValue) : 0); |
3853 | |
|
3854 | 0 | pszValue = CSLFetchNameValue(papszParams, "nodata"); |
3855 | 0 | poPowerOpts->dfNoDataValue = pszValue ? CPLAtofM(pszValue) : 0.0; |
3856 | |
|
3857 | 0 | static const char *const apszKnownOptions[] = { |
3858 | 0 | "power", "smoothing", "radius", "radius1", "radius2", |
3859 | 0 | "angle", "max_points", "min_points", "nodata", nullptr}; |
3860 | 0 | papszKnownOptions = apszKnownOptions; |
3861 | |
|
3862 | 0 | break; |
3863 | 0 | } |
3864 | 0 | case GGA_InverseDistanceToAPowerNearestNeighbor: |
3865 | 0 | { |
3866 | 0 | *ppOptions = CPLMalloc( |
3867 | 0 | sizeof(GDALGridInverseDistanceToAPowerNearestNeighborOptions)); |
3868 | |
|
3869 | 0 | GDALGridInverseDistanceToAPowerNearestNeighborOptions |
3870 | 0 | *const poPowerOpts = static_cast< |
3871 | 0 | GDALGridInverseDistanceToAPowerNearestNeighborOptions *>( |
3872 | 0 | *ppOptions); |
3873 | |
|
3874 | 0 | poPowerOpts->nSizeOfStructure = sizeof(*poPowerOpts); |
3875 | |
|
3876 | 0 | const char *pszValue = CSLFetchNameValue(papszParams, "power"); |
3877 | 0 | poPowerOpts->dfPower = pszValue ? CPLAtofM(pszValue) : 2.0; |
3878 | |
|
3879 | 0 | pszValue = CSLFetchNameValue(papszParams, "smoothing"); |
3880 | 0 | poPowerOpts->dfSmoothing = pszValue ? CPLAtofM(pszValue) : 0.0; |
3881 | |
|
3882 | 0 | pszValue = CSLFetchNameValue(papszParams, "radius"); |
3883 | 0 | poPowerOpts->dfRadius = pszValue ? CPLAtofM(pszValue) : 1.0; |
3884 | 0 | if (!(poPowerOpts->dfRadius > 0)) |
3885 | 0 | { |
3886 | 0 | CPLError(CE_Failure, CPLE_IllegalArg, |
3887 | 0 | "Radius value should be strictly positive"); |
3888 | 0 | CSLDestroy(papszParams); |
3889 | 0 | return CE_Failure; |
3890 | 0 | } |
3891 | | |
3892 | 0 | pszValue = CSLFetchNameValue(papszParams, "max_points"); |
3893 | 0 | poPowerOpts->nMaxPoints = |
3894 | 0 | static_cast<GUInt32>(pszValue ? CPLAtofM(pszValue) : 12); |
3895 | |
|
3896 | 0 | pszValue = CSLFetchNameValue(papszParams, "min_points"); |
3897 | 0 | poPowerOpts->nMinPoints = |
3898 | 0 | static_cast<GUInt32>(pszValue ? CPLAtofM(pszValue) : 0); |
3899 | |
|
3900 | 0 | pszValue = CSLFetchNameValue(papszParams, "nodata"); |
3901 | 0 | poPowerOpts->dfNoDataValue = pszValue ? CPLAtofM(pszValue) : 0.0; |
3902 | |
|
3903 | 0 | pszValue = |
3904 | 0 | CSLFetchNameValue(papszParams, "min_points_per_quadrant"); |
3905 | 0 | poPowerOpts->nMinPointsPerQuadrant = |
3906 | 0 | static_cast<GUInt32>(pszValue ? CPLAtofM(pszValue) : 0); |
3907 | |
|
3908 | 0 | pszValue = |
3909 | 0 | CSLFetchNameValue(papszParams, "max_points_per_quadrant"); |
3910 | 0 | poPowerOpts->nMaxPointsPerQuadrant = |
3911 | 0 | static_cast<GUInt32>(pszValue ? CPLAtofM(pszValue) : 0); |
3912 | |
|
3913 | 0 | static const char *const apszKnownOptions[] = { |
3914 | 0 | "power", |
3915 | 0 | "smoothing", |
3916 | 0 | "radius", |
3917 | 0 | "max_points", |
3918 | 0 | "min_points", |
3919 | 0 | "nodata", |
3920 | 0 | "min_points_per_quadrant", |
3921 | 0 | "max_points_per_quadrant", |
3922 | 0 | nullptr}; |
3923 | 0 | papszKnownOptions = apszKnownOptions; |
3924 | |
|
3925 | 0 | break; |
3926 | 0 | } |
3927 | 0 | case GGA_MovingAverage: |
3928 | 0 | { |
3929 | 0 | *ppOptions = CPLMalloc(sizeof(GDALGridMovingAverageOptions)); |
3930 | |
|
3931 | 0 | GDALGridMovingAverageOptions *const poAverageOpts = |
3932 | 0 | static_cast<GDALGridMovingAverageOptions *>(*ppOptions); |
3933 | |
|
3934 | 0 | poAverageOpts->nSizeOfStructure = sizeof(*poAverageOpts); |
3935 | |
|
3936 | 0 | const char *pszValue = CSLFetchNameValue(papszParams, "radius"); |
3937 | 0 | if (pszValue) |
3938 | 0 | { |
3939 | 0 | poAverageOpts->dfRadius1 = CPLAtofM(pszValue); |
3940 | 0 | poAverageOpts->dfRadius2 = poAverageOpts->dfRadius1; |
3941 | 0 | } |
3942 | 0 | else |
3943 | 0 | { |
3944 | 0 | pszValue = CSLFetchNameValue(papszParams, "radius1"); |
3945 | 0 | poAverageOpts->dfRadius1 = pszValue ? CPLAtofM(pszValue) : 0.0; |
3946 | |
|
3947 | 0 | pszValue = CSLFetchNameValue(papszParams, "radius2"); |
3948 | 0 | poAverageOpts->dfRadius2 = pszValue ? CPLAtofM(pszValue) : 0.0; |
3949 | 0 | } |
3950 | |
|
3951 | 0 | pszValue = CSLFetchNameValue(papszParams, "angle"); |
3952 | 0 | poAverageOpts->dfAngle = pszValue ? CPLAtofM(pszValue) : 0.0; |
3953 | |
|
3954 | 0 | pszValue = CSLFetchNameValue(papszParams, "min_points"); |
3955 | 0 | poAverageOpts->nMinPoints = |
3956 | 0 | static_cast<GUInt32>(pszValue ? CPLAtofM(pszValue) : 0); |
3957 | |
|
3958 | 0 | pszValue = CSLFetchNameValue(papszParams, "max_points"); |
3959 | 0 | poAverageOpts->nMaxPoints = |
3960 | 0 | static_cast<GUInt32>(pszValue ? CPLAtofM(pszValue) : 0); |
3961 | |
|
3962 | 0 | pszValue = CSLFetchNameValue(papszParams, "nodata"); |
3963 | 0 | poAverageOpts->dfNoDataValue = pszValue ? CPLAtofM(pszValue) : 0.0; |
3964 | |
|
3965 | 0 | pszValue = |
3966 | 0 | CSLFetchNameValue(papszParams, "min_points_per_quadrant"); |
3967 | 0 | poAverageOpts->nMinPointsPerQuadrant = |
3968 | 0 | static_cast<GUInt32>(pszValue ? CPLAtofM(pszValue) : 0); |
3969 | |
|
3970 | 0 | pszValue = |
3971 | 0 | CSLFetchNameValue(papszParams, "max_points_per_quadrant"); |
3972 | 0 | poAverageOpts->nMaxPointsPerQuadrant = |
3973 | 0 | static_cast<GUInt32>(pszValue ? CPLAtofM(pszValue) : 0); |
3974 | |
|
3975 | 0 | if (poAverageOpts->nMinPointsPerQuadrant != 0 || |
3976 | 0 | poAverageOpts->nMaxPointsPerQuadrant != 0) |
3977 | 0 | { |
3978 | 0 | if (!(poAverageOpts->dfRadius1 > 0) || |
3979 | 0 | !(poAverageOpts->dfRadius2 > 0)) |
3980 | 0 | { |
3981 | 0 | CPLError(CE_Failure, CPLE_IllegalArg, |
3982 | 0 | "Radius value should be strictly positive when " |
3983 | 0 | "per quadrant parameters are specified"); |
3984 | 0 | CSLDestroy(papszParams); |
3985 | 0 | return CE_Failure; |
3986 | 0 | } |
3987 | 0 | if (poAverageOpts->dfAngle != 0) |
3988 | 0 | { |
3989 | 0 | CPLError(CE_Failure, CPLE_NotSupported, |
3990 | 0 | "angle != 0 not supported when " |
3991 | 0 | "per quadrant parameters are specified"); |
3992 | 0 | CSLDestroy(papszParams); |
3993 | 0 | return CE_Failure; |
3994 | 0 | } |
3995 | 0 | } |
3996 | 0 | else if (poAverageOpts->nMaxPoints > 0) |
3997 | 0 | { |
3998 | 0 | CPLError(CE_Warning, CPLE_AppDefined, |
3999 | 0 | "max_points is ignored unless one of " |
4000 | 0 | "min_points_per_quadrant or max_points_per_quadrant " |
4001 | 0 | "is >= 1"); |
4002 | 0 | } |
4003 | | |
4004 | 0 | static const char *const apszKnownOptions[] = { |
4005 | 0 | "radius", |
4006 | 0 | "radius1", |
4007 | 0 | "radius2", |
4008 | 0 | "angle", |
4009 | 0 | "min_points", |
4010 | 0 | "max_points", |
4011 | 0 | "nodata", |
4012 | 0 | "min_points_per_quadrant", |
4013 | 0 | "max_points_per_quadrant", |
4014 | 0 | nullptr}; |
4015 | 0 | papszKnownOptions = apszKnownOptions; |
4016 | |
|
4017 | 0 | break; |
4018 | 0 | } |
4019 | 0 | case GGA_NearestNeighbor: |
4020 | 0 | { |
4021 | 0 | *ppOptions = CPLMalloc(sizeof(GDALGridNearestNeighborOptions)); |
4022 | |
|
4023 | 0 | GDALGridNearestNeighborOptions *const poNeighborOpts = |
4024 | 0 | static_cast<GDALGridNearestNeighborOptions *>(*ppOptions); |
4025 | |
|
4026 | 0 | poNeighborOpts->nSizeOfStructure = sizeof(*poNeighborOpts); |
4027 | |
|
4028 | 0 | const char *pszValue = CSLFetchNameValue(papszParams, "radius"); |
4029 | 0 | if (pszValue) |
4030 | 0 | { |
4031 | 0 | poNeighborOpts->dfRadius1 = CPLAtofM(pszValue); |
4032 | 0 | poNeighborOpts->dfRadius2 = poNeighborOpts->dfRadius1; |
4033 | 0 | } |
4034 | 0 | else |
4035 | 0 | { |
4036 | 0 | pszValue = CSLFetchNameValue(papszParams, "radius1"); |
4037 | 0 | poNeighborOpts->dfRadius1 = pszValue ? CPLAtofM(pszValue) : 0.0; |
4038 | |
|
4039 | 0 | pszValue = CSLFetchNameValue(papszParams, "radius2"); |
4040 | 0 | poNeighborOpts->dfRadius2 = pszValue ? CPLAtofM(pszValue) : 0.0; |
4041 | 0 | } |
4042 | |
|
4043 | 0 | pszValue = CSLFetchNameValue(papszParams, "angle"); |
4044 | 0 | poNeighborOpts->dfAngle = pszValue ? CPLAtofM(pszValue) : 0.0; |
4045 | |
|
4046 | 0 | pszValue = CSLFetchNameValue(papszParams, "nodata"); |
4047 | 0 | poNeighborOpts->dfNoDataValue = pszValue ? CPLAtofM(pszValue) : 0.0; |
4048 | |
|
4049 | 0 | static const char *const apszKnownOptions[] = { |
4050 | 0 | "radius", "radius1", "radius2", "angle", "nodata", nullptr}; |
4051 | 0 | papszKnownOptions = apszKnownOptions; |
4052 | |
|
4053 | 0 | break; |
4054 | 0 | } |
4055 | 0 | case GGA_MetricMinimum: |
4056 | 0 | case GGA_MetricMaximum: |
4057 | 0 | case GGA_MetricRange: |
4058 | 0 | case GGA_MetricCount: |
4059 | 0 | case GGA_MetricAverageDistance: |
4060 | 0 | case GGA_MetricAverageDistancePts: |
4061 | 0 | { |
4062 | 0 | *ppOptions = CPLMalloc(sizeof(GDALGridDataMetricsOptions)); |
4063 | |
|
4064 | 0 | GDALGridDataMetricsOptions *const poMetricsOptions = |
4065 | 0 | static_cast<GDALGridDataMetricsOptions *>(*ppOptions); |
4066 | |
|
4067 | 0 | poMetricsOptions->nSizeOfStructure = sizeof(*poMetricsOptions); |
4068 | |
|
4069 | 0 | const char *pszValue = CSLFetchNameValue(papszParams, "radius"); |
4070 | 0 | if (pszValue) |
4071 | 0 | { |
4072 | 0 | poMetricsOptions->dfRadius1 = CPLAtofM(pszValue); |
4073 | 0 | poMetricsOptions->dfRadius2 = poMetricsOptions->dfRadius1; |
4074 | 0 | } |
4075 | 0 | else |
4076 | 0 | { |
4077 | 0 | pszValue = CSLFetchNameValue(papszParams, "radius1"); |
4078 | 0 | poMetricsOptions->dfRadius1 = |
4079 | 0 | pszValue ? CPLAtofM(pszValue) : 0.0; |
4080 | |
|
4081 | 0 | pszValue = CSLFetchNameValue(papszParams, "radius2"); |
4082 | 0 | poMetricsOptions->dfRadius2 = |
4083 | 0 | pszValue ? CPLAtofM(pszValue) : 0.0; |
4084 | 0 | } |
4085 | |
|
4086 | 0 | pszValue = CSLFetchNameValue(papszParams, "angle"); |
4087 | 0 | poMetricsOptions->dfAngle = pszValue ? CPLAtofM(pszValue) : 0.0; |
4088 | |
|
4089 | 0 | pszValue = CSLFetchNameValue(papszParams, "min_points"); |
4090 | 0 | poMetricsOptions->nMinPoints = pszValue ? atoi(pszValue) : 0; |
4091 | |
|
4092 | 0 | pszValue = CSLFetchNameValue(papszParams, "nodata"); |
4093 | 0 | poMetricsOptions->dfNoDataValue = |
4094 | 0 | pszValue ? CPLAtofM(pszValue) : 0.0; |
4095 | |
|
4096 | 0 | pszValue = |
4097 | 0 | CSLFetchNameValue(papszParams, "min_points_per_quadrant"); |
4098 | 0 | poMetricsOptions->nMinPointsPerQuadrant = |
4099 | 0 | static_cast<GUInt32>(pszValue ? CPLAtofM(pszValue) : 0); |
4100 | |
|
4101 | 0 | pszValue = |
4102 | 0 | CSLFetchNameValue(papszParams, "max_points_per_quadrant"); |
4103 | 0 | poMetricsOptions->nMaxPointsPerQuadrant = |
4104 | 0 | static_cast<GUInt32>(pszValue ? CPLAtofM(pszValue) : 0); |
4105 | |
|
4106 | 0 | if (poMetricsOptions->nMinPointsPerQuadrant != 0 || |
4107 | 0 | poMetricsOptions->nMaxPointsPerQuadrant != 0) |
4108 | 0 | { |
4109 | 0 | if (*peAlgorithm == GGA_MetricAverageDistancePts) |
4110 | 0 | { |
4111 | 0 | CPLError(CE_Failure, CPLE_NotSupported, |
4112 | 0 | "Algorithm %s not supported when " |
4113 | 0 | "per quadrant parameters are specified", |
4114 | 0 | szAlgNameAverageDistancePts); |
4115 | 0 | CSLDestroy(papszParams); |
4116 | 0 | return CE_Failure; |
4117 | 0 | } |
4118 | 0 | if (!(poMetricsOptions->dfRadius1 > 0) || |
4119 | 0 | !(poMetricsOptions->dfRadius2 > 0)) |
4120 | 0 | { |
4121 | 0 | CPLError(CE_Failure, CPLE_IllegalArg, |
4122 | 0 | "Radius value should be strictly positive when " |
4123 | 0 | "per quadrant parameters are specified"); |
4124 | 0 | CSLDestroy(papszParams); |
4125 | 0 | return CE_Failure; |
4126 | 0 | } |
4127 | 0 | if (poMetricsOptions->dfAngle != 0) |
4128 | 0 | { |
4129 | 0 | CPLError(CE_Failure, CPLE_NotSupported, |
4130 | 0 | "angle != 0 not supported when " |
4131 | 0 | "per quadrant parameters are specified"); |
4132 | 0 | CSLDestroy(papszParams); |
4133 | 0 | return CE_Failure; |
4134 | 0 | } |
4135 | 0 | } |
4136 | | |
4137 | 0 | static const char *const apszKnownOptions[] = { |
4138 | 0 | "radius", |
4139 | 0 | "radius1", |
4140 | 0 | "radius2", |
4141 | 0 | "angle", |
4142 | 0 | "min_points", |
4143 | 0 | "nodata", |
4144 | 0 | "min_points_per_quadrant", |
4145 | 0 | "max_points_per_quadrant", |
4146 | 0 | nullptr}; |
4147 | 0 | papszKnownOptions = apszKnownOptions; |
4148 | |
|
4149 | 0 | break; |
4150 | 0 | } |
4151 | 0 | case GGA_Linear: |
4152 | 0 | { |
4153 | 0 | *ppOptions = CPLMalloc(sizeof(GDALGridLinearOptions)); |
4154 | |
|
4155 | 0 | GDALGridLinearOptions *const poLinearOpts = |
4156 | 0 | static_cast<GDALGridLinearOptions *>(*ppOptions); |
4157 | |
|
4158 | 0 | poLinearOpts->nSizeOfStructure = sizeof(*poLinearOpts); |
4159 | |
|
4160 | 0 | const char *pszValue = CSLFetchNameValue(papszParams, "radius"); |
4161 | 0 | poLinearOpts->dfRadius = pszValue ? CPLAtofM(pszValue) : -1.0; |
4162 | |
|
4163 | 0 | pszValue = CSLFetchNameValue(papszParams, "nodata"); |
4164 | 0 | poLinearOpts->dfNoDataValue = pszValue ? CPLAtofM(pszValue) : 0.0; |
4165 | |
|
4166 | 0 | static const char *const apszKnownOptions[] = {"radius", "nodata", |
4167 | 0 | nullptr}; |
4168 | 0 | papszKnownOptions = apszKnownOptions; |
4169 | |
|
4170 | 0 | break; |
4171 | 0 | } |
4172 | 0 | } |
4173 | | |
4174 | 0 | if (papszKnownOptions) |
4175 | 0 | { |
4176 | 0 | for (int i = 1; papszParams[i] != nullptr; ++i) |
4177 | 0 | { |
4178 | 0 | char *pszKey = nullptr; |
4179 | 0 | CPL_IGNORE_RET_VAL(CPLParseNameValue(papszParams[i], &pszKey)); |
4180 | 0 | if (pszKey) |
4181 | 0 | { |
4182 | 0 | bool bKnownKey = false; |
4183 | 0 | for (const char *const *papszKnownKeyIter = papszKnownOptions; |
4184 | 0 | *papszKnownKeyIter; ++papszKnownKeyIter) |
4185 | 0 | { |
4186 | 0 | if (EQUAL(*papszKnownKeyIter, pszKey)) |
4187 | 0 | { |
4188 | 0 | bKnownKey = true; |
4189 | 0 | break; |
4190 | 0 | } |
4191 | 0 | } |
4192 | 0 | if (!bKnownKey) |
4193 | 0 | { |
4194 | 0 | CPLError(CE_Warning, CPLE_AppDefined, "Option %s ignored", |
4195 | 0 | pszKey); |
4196 | 0 | } |
4197 | 0 | } |
4198 | 0 | CPLFree(pszKey); |
4199 | 0 | } |
4200 | 0 | } |
4201 | |
|
4202 | 0 | CSLDestroy(papszParams); |
4203 | 0 | return CE_None; |
4204 | 0 | } |