compositor: bokeh blur size input can now be an image, in this case it uses VariableS...
[blender-staging.git] / source / blender / compositor / operations / COM_VariableSizeBokehBlurOperation.cpp
1 /*
2  * Copyright 2011, Blender Foundation.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public License
6  * as published by the Free Software Foundation; either version 2
7  * of the License, or (at your option) any later version.
8  *
9  * This program is distributed in the hope that it will be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12  * GNU General Public License for more details.
13  *
14  * You should have received a copy of the GNU General Public License
15  * along with this program; if not, write to the Free Software Foundation,
16  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
17  *
18  * Contributor: 
19  *              Jeroen Bakker 
20  *              Monique Dewanchand
21  */
22
23 #include "COM_VariableSizeBokehBlurOperation.h"
24 #include "BLI_math.h"
25 #include "COM_OpenCLDevice.h"
26
27 extern "C" {
28         #include "RE_pipeline.h"
29 }
30
31 VariableSizeBokehBlurOperation::VariableSizeBokehBlurOperation() : NodeOperation()
32 {
33         this->addInputSocket(COM_DT_COLOR);
34         this->addInputSocket(COM_DT_COLOR, COM_SC_NO_RESIZE); // do not resize the bokeh image.
35         this->addInputSocket(COM_DT_VALUE); // radius
36 #ifdef COM_DEFOCUS_SEARCH
37         this->addInputSocket(COM_DT_COLOR, COM_SC_NO_RESIZE); // inverse search radius optimization structure.
38 #endif
39         this->addOutputSocket(COM_DT_COLOR);
40         this->setComplex(true);
41         this->setOpenCL(true);
42
43         this->m_inputProgram = NULL;
44         this->m_inputBokehProgram = NULL;
45         this->m_inputSizeProgram = NULL;
46         this->m_maxBlur = 32.0f;
47         this->m_threshold = 1.0f;
48         this->m_do_size_scale = false;
49 #ifdef COM_DEFOCUS_SEARCH
50         this->m_inputSearchProgram = NULL;
51 #endif
52 }
53
54
55 void VariableSizeBokehBlurOperation::initExecution()
56 {
57         this->m_inputProgram = getInputSocketReader(0);
58         this->m_inputBokehProgram = getInputSocketReader(1);
59         this->m_inputSizeProgram = getInputSocketReader(2);
60 #ifdef COM_DEFOCUS_SEARCH
61         this->m_inputSearchProgram = getInputSocketReader(3);
62 #endif
63         QualityStepHelper::initExecution(COM_QH_INCREASE);
64 }
65 struct VariableSizeBokehBlurTileData
66 {
67         MemoryBuffer *color;
68         MemoryBuffer *bokeh;
69         MemoryBuffer *size;
70         int maxBlurScalar;
71 };
72
73 void *VariableSizeBokehBlurOperation::initializeTileData(rcti *rect)
74 {
75         VariableSizeBokehBlurTileData *data = new VariableSizeBokehBlurTileData();
76         data->color = (MemoryBuffer *)this->m_inputProgram->initializeTileData(rect);
77         data->bokeh = (MemoryBuffer *)this->m_inputBokehProgram->initializeTileData(rect);
78         data->size = (MemoryBuffer *)this->m_inputSizeProgram->initializeTileData(rect);
79
80
81         rcti rect2;
82         this->determineDependingAreaOfInterest(rect, (ReadBufferOperation *)this->m_inputSizeProgram, &rect2);
83
84         const float max_dim = max(m_width, m_height);
85         const float scalar = this->m_do_size_scale ? (max_dim / 100.0f) : 1.0f;
86
87         data->maxBlurScalar = (int)(data->size->getMaximumValue(&rect2) * scalar);
88         CLAMP(data->maxBlurScalar, 1.0f, this->m_maxBlur);
89         return data;
90 }
91
92 void VariableSizeBokehBlurOperation::deinitializeTileData(rcti *rect, void *data)
93 {
94         VariableSizeBokehBlurTileData *result = (VariableSizeBokehBlurTileData *)data;
95         delete result;
96 }
97
98 void VariableSizeBokehBlurOperation::executePixel(float output[4], int x, int y, void *data)
99 {
100         VariableSizeBokehBlurTileData *tileData = (VariableSizeBokehBlurTileData *)data;
101         MemoryBuffer *inputProgramBuffer = tileData->color;
102         MemoryBuffer *inputBokehBuffer = tileData->bokeh;
103         MemoryBuffer *inputSizeBuffer = tileData->size;
104         float *inputSizeFloatBuffer = inputSizeBuffer->getBuffer();
105         float *inputProgramFloatBuffer = inputProgramBuffer->getBuffer();
106         float readColor[4];
107         float bokeh[4];
108         float tempSize[4];
109         float multiplier_accum[4];
110         float color_accum[4];
111
112         const float max_dim = max(m_width, m_height);
113         const float scalar = this->m_do_size_scale ? (max_dim / 100.0f) : 1.0f;
114         int maxBlurScalar = tileData->maxBlurScalar;
115
116         BLI_assert(inputBokehBuffer->getWidth()  == COM_BLUR_BOKEH_PIXELS);
117         BLI_assert(inputBokehBuffer->getHeight() == COM_BLUR_BOKEH_PIXELS);
118
119 #ifdef COM_DEFOCUS_SEARCH
120         float search[4];
121         this->m_inputSearchProgram->read(search, x/InverseSearchRadiusOperation::DIVIDER, y / InverseSearchRadiusOperation::DIVIDER, NULL);
122         int minx = search[0];
123         int miny = search[1];
124         int maxx = search[2];
125         int maxy = search[3];
126 #else
127         int minx = max(x - maxBlurScalar, 0);
128         int miny = max(y - maxBlurScalar, 0);
129         int maxx = min(x + maxBlurScalar, (int)m_width);
130         int maxy = min(y + maxBlurScalar, (int)m_height);
131 #endif
132         {
133                 inputSizeBuffer->readNoCheck(tempSize, x, y);
134                 inputProgramBuffer->readNoCheck(readColor, x, y);
135
136                 copy_v4_v4(color_accum, readColor);
137                 copy_v4_fl(multiplier_accum, 1.0f);
138                 float size_center = tempSize[0] * scalar;
139                 
140                 const int addXStep = QualityStepHelper::getStep() * COM_NUMBER_OF_CHANNELS;
141                 
142                 if (size_center > this->m_threshold) {
143                         for (int ny = miny; ny < maxy; ny += QualityStepHelper::getStep()) {
144                                 float dy = ny - y;
145                                 int offsetNy = ny * inputSizeBuffer->getWidth() * COM_NUMBER_OF_CHANNELS;
146                                 int offsetNxNy = offsetNy + (minx * COM_NUMBER_OF_CHANNELS);
147                                 for (int nx = minx; nx < maxx; nx += QualityStepHelper::getStep()) {
148                                         if (nx != x || ny != y) {
149                                                 float size = inputSizeFloatBuffer[offsetNxNy] * scalar;
150                                                 if (size > this->m_threshold) {
151                                                         float dx = nx - x;
152                                                         if (size > fabsf(dx) && size > fabsf(dy)) {
153                                                                 float uv[2] = {
154                                                                     (float)(COM_BLUR_BOKEH_PIXELS / 2) + (dx / size) * (float)((COM_BLUR_BOKEH_PIXELS / 2) - 1),
155                                                                     (float)(COM_BLUR_BOKEH_PIXELS / 2) + (dy / size) * (float)((COM_BLUR_BOKEH_PIXELS / 2) - 1)};
156                                                                 inputBokehBuffer->readNoCheck(bokeh, uv[0], uv[1]);
157                                                                 madd_v4_v4v4(color_accum, bokeh, &inputProgramFloatBuffer[offsetNxNy]);
158                                                                 add_v4_v4(multiplier_accum, bokeh);
159                                                         }
160                                                 }
161                                         }
162                                         offsetNxNy += addXStep;
163                                 }
164                         }
165                 }
166
167                 output[0] = color_accum[0] / multiplier_accum[0];
168                 output[1] = color_accum[1] / multiplier_accum[1];
169                 output[2] = color_accum[2] / multiplier_accum[2];
170                 output[3] = color_accum[3] / multiplier_accum[3];
171
172                 /* blend in out values over the threshold, otherwise we get sharp, ugly transitions */
173                 if ((size_center > this->m_threshold) &&
174                     (size_center < this->m_threshold * 2.0f))
175                 {
176                         /* factor from 0-1 */
177                         float fac = (size_center - this->m_threshold) / this->m_threshold;
178                         interp_v4_v4v4(output, readColor, output, fac);
179                 }
180         }
181
182 }
183
184 void VariableSizeBokehBlurOperation::executeOpenCL(OpenCLDevice *device,
185                                        MemoryBuffer *outputMemoryBuffer, cl_mem clOutputBuffer, 
186                                        MemoryBuffer **inputMemoryBuffers, list<cl_mem> *clMemToCleanUp, 
187                                        list<cl_kernel> *clKernelsToCleanUp) 
188 {
189         cl_kernel defocusKernel = device->COM_clCreateKernel("defocusKernel", NULL);
190
191         cl_int step = this->getStep();
192         cl_int maxBlur;
193         cl_float threshold = this->m_threshold;
194         
195         MemoryBuffer *sizeMemoryBuffer = (MemoryBuffer *)this->m_inputSizeProgram->getInputMemoryBuffer(inputMemoryBuffers);
196
197         const float max_dim = max(m_width, m_height);
198         cl_float scalar = this->m_do_size_scale ? (max_dim / 100.0f) : 1.0f;
199
200         maxBlur = (cl_int)sizeMemoryBuffer->getMaximumValue() * scalar;
201         maxBlur = MIN2(maxBlur, this->m_maxBlur);
202
203         device->COM_clAttachMemoryBufferToKernelParameter(defocusKernel, 0, -1, clMemToCleanUp, inputMemoryBuffers, this->m_inputProgram);
204         device->COM_clAttachMemoryBufferToKernelParameter(defocusKernel, 1,  -1, clMemToCleanUp, inputMemoryBuffers, this->m_inputBokehProgram);
205         device->COM_clAttachMemoryBufferToKernelParameter(defocusKernel, 2,  4, clMemToCleanUp, inputMemoryBuffers, this->m_inputSizeProgram);
206         device->COM_clAttachOutputMemoryBufferToKernelParameter(defocusKernel, 3, clOutputBuffer);
207         device->COM_clAttachMemoryBufferOffsetToKernelParameter(defocusKernel, 5, outputMemoryBuffer);
208         clSetKernelArg(defocusKernel, 6, sizeof(cl_int), &step);
209         clSetKernelArg(defocusKernel, 7, sizeof(cl_int), &maxBlur);
210         clSetKernelArg(defocusKernel, 8, sizeof(cl_float), &threshold);
211         clSetKernelArg(defocusKernel, 9, sizeof(cl_float), &scalar);
212         device->COM_clAttachSizeToKernelParameter(defocusKernel, 10, this);
213         
214         device->COM_clEnqueueRange(defocusKernel, outputMemoryBuffer, 11, this);
215 }
216
217 void VariableSizeBokehBlurOperation::deinitExecution()
218 {
219         this->m_inputProgram = NULL;
220         this->m_inputBokehProgram = NULL;
221         this->m_inputSizeProgram = NULL;
222 #ifdef COM_DEFOCUS_SEARCH
223         this->m_inputSearchProgram = NULL;
224 #endif
225 }
226
227 bool VariableSizeBokehBlurOperation::determineDependingAreaOfInterest(rcti *input, ReadBufferOperation *readOperation, rcti *output)
228 {
229         rcti newInput;
230         rcti bokehInput;
231
232         const float max_dim = max(m_width, m_height);
233         const float scalar = this->m_do_size_scale ? (max_dim / 100.0f) : 1.0f;
234         int maxBlurScalar = this->m_maxBlur * scalar;
235
236         newInput.xmax = input->xmax + maxBlurScalar + 2;
237         newInput.xmin = input->xmin - maxBlurScalar + 2;
238         newInput.ymax = input->ymax + maxBlurScalar - 2;
239         newInput.ymin = input->ymin - maxBlurScalar - 2;
240         bokehInput.xmax = COM_BLUR_BOKEH_PIXELS;
241         bokehInput.xmin = 0;
242         bokehInput.ymax = COM_BLUR_BOKEH_PIXELS;
243         bokehInput.ymin = 0;
244         
245
246         NodeOperation *operation = getInputOperation(2);
247         if (operation->determineDependingAreaOfInterest(&newInput, readOperation, output) ) {
248                 return true;
249         }
250         operation = getInputOperation(1);
251         if (operation->determineDependingAreaOfInterest(&bokehInput, readOperation, output) ) {
252                 return true;
253         }
254 #ifdef COM_DEFOCUS_SEARCH
255         rcti searchInput;
256         searchInput.xmax = (input->xmax / InverseSearchRadiusOperation::DIVIDER) + 1;
257         searchInput.xmin = (input->xmin / InverseSearchRadiusOperation::DIVIDER) - 1;
258         searchInput.ymax = (input->ymax / InverseSearchRadiusOperation::DIVIDER) + 1;
259         searchInput.ymin = (input->ymin / InverseSearchRadiusOperation::DIVIDER) - 1;
260         operation = getInputOperation(3);
261         if (operation->determineDependingAreaOfInterest(&searchInput, readOperation, output) ) {
262                 return true;
263         }
264 #endif
265         operation = getInputOperation(0);
266         if (operation->determineDependingAreaOfInterest(&newInput, readOperation, output) ) {
267                 return true;
268         }
269         return false;
270 }
271
272 #ifdef COM_DEFOCUS_SEARCH
273 // InverseSearchRadiusOperation
274 InverseSearchRadiusOperation::InverseSearchRadiusOperation() : NodeOperation() 
275 {
276         this->addInputSocket(COM_DT_VALUE, COM_SC_NO_RESIZE); // radius
277         this->addOutputSocket(COM_DT_COLOR);
278         this->setComplex(true);
279         this->m_inputRadius = NULL;
280 }
281
282 void InverseSearchRadiusOperation::initExecution() 
283 {
284         this->m_inputRadius = this->getInputSocketReader(0);
285 }
286
287 voi *InverseSearchRadiusOperation::initializeTileData(rcti *rect)
288 {
289         MemoryBuffer * data = new MemoryBuffer(NULL, rect);
290         float *buffer = data->getBuffer();
291         int x, y;
292         int width = this->m_inputRadius->getWidth();
293         int height = this->m_inputRadius->getHeight();
294         float temp[4];
295         int offset = 0;
296         for (y = rect->ymin; y < rect->ymax ; y++) {
297                 for (x = rect->xmin; x < rect->xmax ; x++) {
298                         int rx = x * DIVIDER;
299                         int ry = y * DIVIDER;
300                         buffer[offset] = MAX2(rx - m_maxBlur, 0);
301                         buffer[offset+1] = MAX2(ry- m_maxBlur, 0);
302                         buffer[offset+2] = MIN2(rx+DIVIDER + m_maxBlur, width);
303                         buffer[offset+3] = MIN2(ry+DIVIDER + m_maxBlur, height);
304                         offset += 4;
305                 }
306         }
307 //      for (x = rect->xmin; x < rect->xmax ; x++) {
308 //              for (y = rect->ymin; y < rect->ymax ; y++) {
309 //                      int rx = x * DIVIDER;
310 //                      int ry = y * DIVIDER;
311 //                      float radius = 0.0f;
312 //                      float maxx = x;
313 //                      float maxy = y;
314         
315 //                      for (int x2 = 0 ; x2 < DIVIDER ; x2 ++) {
316 //                              for (int y2 = 0 ; y2 < DIVIDER ; y2 ++) {
317 //                                      this->m_inputRadius->read(temp, rx+x2, ry+y2, COM_PS_NEAREST);
318 //                                      if (radius < temp[0]) {
319 //                                              radius = temp[0];
320 //                                              maxx = x2;
321 //                                              maxy = y2;
322 //                                      }
323 //                              }
324 //                      }
325 //                      int impactRadius = ceil(radius / DIVIDER);
326 //                      for (int x2 = x - impactRadius ; x2 < x + impactRadius ; x2 ++) {
327 //                              for (int y2 = y - impactRadius ; y2 < y + impactRadius ; y2 ++) {
328 //                                      data->read(temp, x2, y2);
329 //                                      temp[0] = MIN2(temp[0], maxx);
330 //                                      temp[1] = MIN2(temp[1], maxy);
331 //                                      temp[2] = MAX2(temp[2], maxx);
332 //                                      temp[3] = MAX2(temp[3], maxy);
333 //                                      data->writePixel(x2, y2, temp);
334 //                              }
335 //                      }
336 //              }
337 //      }
338         return data;
339 }
340
341 void InverseSearchRadiusOperation::executePixel(float output[4], int x, int y, void *data)
342 {
343         MemoryBuffer *buffer = (MemoryBuffer *)data;
344         buffer->readNoCheck(color, x, y);
345 }
346
347 void InverseSearchRadiusOperation::deinitializeTileData(rcti *rect, void *data) 
348 {
349         if (data) {
350                 MemoryBuffer *mb = (MemoryBuffer *)data;
351                 delete mb;
352         }
353 }
354
355 void InverseSearchRadiusOperation::deinitExecution() 
356 {
357         this->m_inputRadius = NULL;
358 }
359
360 void InverseSearchRadiusOperation::determineResolution(unsigned int resolution[2], unsigned int preferredResolution[2])
361 {
362         NodeOperation::determineResolution(resolution, preferredResolution);
363         resolution[0] = resolution[0] / DIVIDER;
364         resolution[1] = resolution[1] / DIVIDER;
365 }
366
367 bool InverseSearchRadiusOperation::determineDependingAreaOfInterest(rcti *input, ReadBufferOperation *readOperation, rcti *output)
368 {
369         rcti newRect;
370         newRect.ymin = input->ymin * DIVIDER - m_maxBlur;
371         newRect.ymax = input->ymax * DIVIDER + m_maxBlur;
372         newRect.xmin = input->xmin * DIVIDER - m_maxBlur;
373         newRect.xmax = input->xmax * DIVIDER + m_maxBlur;
374         return NodeOperation::determineDependingAreaOfInterest(&newRect, readOperation, output);
375 }
376 #endif