Merge branch 'master' into blender2.8
[blender.git] / intern / cycles / render / buffers.cpp
1 /*
2  * Copyright 2011-2013 Blender Foundation
3  *
4  * Licensed under the Apache License, Version 2.0 (the "License");
5  * you may not use this file except in compliance with the License.
6  * You may obtain a copy of the License at
7  *
8  * http://www.apache.org/licenses/LICENSE-2.0
9  *
10  * Unless required by applicable law or agreed to in writing, software
11  * distributed under the License is distributed on an "AS IS" BASIS,
12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13  * See the License for the specific language governing permissions and
14  * limitations under the License.
15  */
16
17 #include <stdlib.h>
18
19 #include "render/buffers.h"
20 #include "device/device.h"
21
22 #include "util/util_debug.h"
23 #include "util/util_foreach.h"
24 #include "util/util_hash.h"
25 #include "util/util_image.h"
26 #include "util/util_math.h"
27 #include "util/util_opengl.h"
28 #include "util/util_time.h"
29 #include "util/util_types.h"
30
31 CCL_NAMESPACE_BEGIN
32
33 /* Buffer Params */
34
35 BufferParams::BufferParams()
36 {
37         width = 0;
38         height = 0;
39
40         full_x = 0;
41         full_y = 0;
42         full_width = 0;
43         full_height = 0;
44
45         denoising_data_pass = false;
46         denoising_clean_pass = false;
47
48         Pass::add(PASS_COMBINED, passes);
49 }
50
51 void BufferParams::get_offset_stride(int& offset, int& stride)
52 {
53         offset = -(full_x + full_y*width);
54         stride = width;
55 }
56
57 bool BufferParams::modified(const BufferParams& params)
58 {
59         return !(full_x == params.full_x
60                 && full_y == params.full_y
61                 && width == params.width
62                 && height == params.height
63                 && full_width == params.full_width
64                 && full_height == params.full_height
65                 && Pass::equals(passes, params.passes));
66 }
67
68 int BufferParams::get_passes_size()
69 {
70         int size = 0;
71
72         for(size_t i = 0; i < passes.size(); i++)
73                 size += passes[i].components;
74
75         if(denoising_data_pass) {
76                 size += DENOISING_PASS_SIZE_BASE;
77                 if(denoising_clean_pass) size += DENOISING_PASS_SIZE_CLEAN;
78         }
79
80         return align_up(size, 4);
81 }
82
83 int BufferParams::get_denoising_offset()
84 {
85         int offset = 0;
86
87         for(size_t i = 0; i < passes.size(); i++)
88                 offset += passes[i].components;
89
90         return offset;
91 }
92
93 /* Render Buffer Task */
94
95 RenderTile::RenderTile()
96 {
97         x = 0;
98         y = 0;
99         w = 0;
100         h = 0;
101
102         sample = 0;
103         start_sample = 0;
104         num_samples = 0;
105         resolution = 0;
106
107         offset = 0;
108         stride = 0;
109
110         buffer = 0;
111
112         buffers = NULL;
113 }
114
115 /* Render Buffers */
116
117 RenderBuffers::RenderBuffers(Device *device)
118 : buffer(device, "RenderBuffers", MEM_READ_WRITE),
119   map_neighbor_copied(false), render_time(0.0f)
120 {
121 }
122
123 RenderBuffers::~RenderBuffers()
124 {
125         buffer.free();
126 }
127
128 void RenderBuffers::reset(BufferParams& params_)
129 {
130         params = params_;
131
132         /* re-allocate buffer */
133         buffer.alloc(params.width*params.height*params.get_passes_size());
134         buffer.zero_to_device();
135 }
136
137 void RenderBuffers::zero()
138 {
139         buffer.zero_to_device();
140 }
141
142 bool RenderBuffers::copy_from_device()
143 {
144         if(!buffer.device_pointer)
145                 return false;
146
147         buffer.copy_from_device(0, params.width * params.get_passes_size(), params.height);
148
149         return true;
150 }
151
152 bool RenderBuffers::get_denoising_pass_rect(int offset, float exposure, int sample, int components, float *pixels)
153 {
154         float invsample = 1.0f/sample;
155         float scale = invsample;
156         bool variance = (offset == DENOISING_PASS_NORMAL_VAR) ||
157                         (offset == DENOISING_PASS_ALBEDO_VAR) ||
158                         (offset == DENOISING_PASS_DEPTH_VAR) ||
159                         (offset == DENOISING_PASS_COLOR_VAR);
160
161         if(offset == DENOISING_PASS_COLOR) {
162                 scale *= exposure;
163         }
164         else if(offset == DENOISING_PASS_COLOR_VAR) {
165                 scale *= exposure*exposure;
166         }
167
168         offset += params.get_denoising_offset();
169         int pass_stride = params.get_passes_size();
170         int size = params.width*params.height;
171
172         if(variance) {
173                 /* Approximate variance as E[x^2] - 1/N * (E[x])^2, since online variance
174                  * update does not work efficiently with atomics in the kernel. */
175                 int mean_offset = offset - components;
176                 float *mean = buffer.data() + mean_offset;
177                 float *var = buffer.data() + offset;
178                 assert(mean_offset >= 0);
179
180                 if(components == 1) {
181                         for(int i = 0; i < size; i++, mean += pass_stride, var += pass_stride, pixels++) {
182                                 pixels[0] = max(0.0f, var[0] - mean[0]*mean[0]*invsample)*scale;
183                         }
184                 }
185                 else if(components == 3) {
186                         for(int i = 0; i < size; i++, mean += pass_stride, var += pass_stride, pixels += 3) {
187                                 pixels[0] = max(0.0f, var[0] - mean[0]*mean[0]*invsample)*scale;
188                                 pixels[1] = max(0.0f, var[1] - mean[1]*mean[1]*invsample)*scale;
189                                 pixels[2] = max(0.0f, var[2] - mean[2]*mean[2]*invsample)*scale;
190                         }
191                 }
192                 else {
193                         return false;
194                 }
195         }
196         else {
197                 float *in = buffer.data() + offset;
198
199                 if(components == 1) {
200                         for(int i = 0; i < size; i++, in += pass_stride, pixels++) {
201                                 pixels[0] = in[0]*scale;
202                         }
203                 }
204                 else if(components == 3) {
205                         for(int i = 0; i < size; i++, in += pass_stride, pixels += 3) {
206                                 pixels[0] = in[0]*scale;
207                                 pixels[1] = in[1]*scale;
208                                 pixels[2] = in[2]*scale;
209                         }
210                 }
211                 else {
212                         return false;
213                 }
214         }
215
216         return true;
217 }
218
219 bool RenderBuffers::get_pass_rect(PassType type, float exposure, int sample, int components, float *pixels)
220 {
221         int pass_offset = 0;
222
223         for(size_t j = 0; j < params.passes.size(); j++) {
224                 Pass& pass = params.passes[j];
225
226                 if(pass.type != type) {
227                         pass_offset += pass.components;
228                         continue;
229                 }
230
231                 float *in = buffer.data() + pass_offset;
232                 int pass_stride = params.get_passes_size();
233
234                 float scale = (pass.filter)? 1.0f/(float)sample: 1.0f;
235                 float scale_exposure = (pass.exposure)? scale*exposure: scale;
236
237                 int size = params.width*params.height;
238
239                 if(components == 1 && type == PASS_RENDER_TIME) {
240                         /* Render time is not stored by kernel, but measured per tile. */
241                         float val = (float) (1000.0 * render_time/(params.width * params.height * sample));
242                         for(int i = 0; i < size; i++, pixels++) {
243                                 pixels[0] = val;
244                         }
245                 }
246                 else if(components == 1) {
247                         assert(pass.components == components);
248
249                         /* Scalar */
250                         if(type == PASS_DEPTH) {
251                                 for(int i = 0; i < size; i++, in += pass_stride, pixels++) {
252                                         float f = *in;
253                                         pixels[0] = (f == 0.0f)? 1e10f: f*scale_exposure;
254                                 }
255                         }
256                         else if(type == PASS_MIST) {
257                                 for(int i = 0; i < size; i++, in += pass_stride, pixels++) {
258                                         float f = *in;
259                                         pixels[0] = saturate(f*scale_exposure);
260                                 }
261                         }
262 #ifdef WITH_CYCLES_DEBUG
263                         else if(type == PASS_BVH_TRAVERSED_NODES ||
264                                 type == PASS_BVH_TRAVERSED_INSTANCES ||
265                                 type == PASS_BVH_INTERSECTIONS ||
266                                 type == PASS_RAY_BOUNCES)
267                         {
268                                 for(int i = 0; i < size; i++, in += pass_stride, pixels++) {
269                                         float f = *in;
270                                         pixels[0] = f*scale;
271                                 }
272                         }
273 #endif
274                         else {
275                                 for(int i = 0; i < size; i++, in += pass_stride, pixels++) {
276                                         float f = *in;
277                                         pixels[0] = f*scale_exposure;
278                                 }
279                         }
280                 }
281                 else if(components == 3) {
282                         assert(pass.components == 4);
283
284                         /* RGBA */
285                         if(type == PASS_SHADOW) {
286                                 for(int i = 0; i < size; i++, in += pass_stride, pixels += 3) {
287                                         float4 f = make_float4(in[0], in[1], in[2], in[3]);
288                                         float invw = (f.w > 0.0f)? 1.0f/f.w: 1.0f;
289
290                                         pixels[0] = f.x*invw;
291                                         pixels[1] = f.y*invw;
292                                         pixels[2] = f.z*invw;
293                                 }
294                         }
295                         else if(pass.divide_type != PASS_NONE) {
296                                 /* RGB lighting passes that need to divide out color */
297                                 pass_offset = 0;
298                                 for(size_t k = 0; k < params.passes.size(); k++) {
299                                         Pass& color_pass = params.passes[k];
300                                         if(color_pass.type == pass.divide_type)
301                                                 break;
302                                         pass_offset += color_pass.components;
303                                 }
304
305                                 float *in_divide = buffer.data() + pass_offset;
306
307                                 for(int i = 0; i < size; i++, in += pass_stride, in_divide += pass_stride, pixels += 3) {
308                                         float3 f = make_float3(in[0], in[1], in[2]);
309                                         float3 f_divide = make_float3(in_divide[0], in_divide[1], in_divide[2]);
310
311                                         f = safe_divide_even_color(f*exposure, f_divide);
312
313                                         pixels[0] = f.x;
314                                         pixels[1] = f.y;
315                                         pixels[2] = f.z;
316                                 }
317                         }
318                         else {
319                                 /* RGB/vector */
320                                 for(int i = 0; i < size; i++, in += pass_stride, pixels += 3) {
321                                         float3 f = make_float3(in[0], in[1], in[2]);
322
323                                         pixels[0] = f.x*scale_exposure;
324                                         pixels[1] = f.y*scale_exposure;
325                                         pixels[2] = f.z*scale_exposure;
326                                 }
327                         }
328                 }
329                 else if(components == 4) {
330                         assert(pass.components == components);
331
332                         /* RGBA */
333                         if(type == PASS_SHADOW) {
334                                 for(int i = 0; i < size; i++, in += pass_stride, pixels += 4) {
335                                         float4 f = make_float4(in[0], in[1], in[2], in[3]);
336                                         float invw = (f.w > 0.0f)? 1.0f/f.w: 1.0f;
337
338                                         pixels[0] = f.x*invw;
339                                         pixels[1] = f.y*invw;
340                                         pixels[2] = f.z*invw;
341                                         pixels[3] = 1.0f;
342                                 }
343                         }
344                         else if(type == PASS_MOTION) {
345                                 /* need to normalize by number of samples accumulated for motion */
346                                 pass_offset = 0;
347                                 for(size_t k = 0; k < params.passes.size(); k++) {
348                                         Pass& color_pass = params.passes[k];
349                                         if(color_pass.type == PASS_MOTION_WEIGHT)
350                                                 break;
351                                         pass_offset += color_pass.components;
352                                 }
353
354                                 float *in_weight = buffer.data() + pass_offset;
355
356                                 for(int i = 0; i < size; i++, in += pass_stride, in_weight += pass_stride, pixels += 4) {
357                                         float4 f = make_float4(in[0], in[1], in[2], in[3]);
358                                         float w = in_weight[0];
359                                         float invw = (w > 0.0f)? 1.0f/w: 0.0f;
360
361                                         pixels[0] = f.x*invw;
362                                         pixels[1] = f.y*invw;
363                                         pixels[2] = f.z*invw;
364                                         pixels[3] = f.w*invw;
365                                 }
366                         }
367                         else {
368                                 for(int i = 0; i < size; i++, in += pass_stride, pixels += 4) {
369                                         float4 f = make_float4(in[0], in[1], in[2], in[3]);
370
371                                         pixels[0] = f.x*scale_exposure;
372                                         pixels[1] = f.y*scale_exposure;
373                                         pixels[2] = f.z*scale_exposure;
374
375                                         /* clamp since alpha might be > 1.0 due to russian roulette */
376                                         pixels[3] = saturate(f.w*scale);
377                                 }
378                         }
379                 }
380
381                 return true;
382         }
383
384         return false;
385 }
386
387 /* Display Buffer */
388
389 DisplayBuffer::DisplayBuffer(Device *device, bool linear)
390 : draw_width(0),
391   draw_height(0),
392   transparent(true), /* todo: determine from background */
393   half_float(linear),
394   rgba_byte(device, "display buffer byte"),
395   rgba_half(device, "display buffer half")
396 {
397 }
398
399 DisplayBuffer::~DisplayBuffer()
400 {
401         rgba_byte.free();
402         rgba_half.free();
403 }
404
405 void DisplayBuffer::reset(BufferParams& params_)
406 {
407         draw_width = 0;
408         draw_height = 0;
409
410         params = params_;
411
412         /* allocate display pixels */
413         if(half_float) {
414                 rgba_half.alloc_to_device(params.width, params.height);
415         }
416         else {
417                 rgba_byte.alloc_to_device(params.width, params.height);
418         }
419 }
420
421 void DisplayBuffer::draw_set(int width, int height)
422 {
423         assert(width <= params.width && height <= params.height);
424
425         draw_width = width;
426         draw_height = height;
427 }
428
429 void DisplayBuffer::draw(Device *device, const DeviceDrawParams& draw_params)
430 {
431         if(draw_width != 0 && draw_height != 0) {
432                 device_memory& rgba = (half_float)? (device_memory&)rgba_half:
433                                                     (device_memory&)rgba_byte;
434
435                 device->draw_pixels(
436                             rgba, 0,
437                             draw_width, draw_height, params.width, params.height,
438                             params.full_x, params.full_y, params.full_width, params.full_height,
439                             transparent, draw_params);
440         }
441 }
442
443 bool DisplayBuffer::draw_ready()
444 {
445         return (draw_width != 0 && draw_height != 0);
446 }
447
448 void DisplayBuffer::write(const string& filename)
449 {
450         int w = draw_width;
451         int h = draw_height;
452
453         if(w == 0 || h == 0)
454                 return;
455         
456         if(half_float)
457                 return;
458
459         /* read buffer from device */
460         uchar4 *pixels = rgba_byte.copy_from_device(0, w, h);
461
462         /* write image */
463         ImageOutput *out = ImageOutput::create(filename);
464         ImageSpec spec(w, h, 4, TypeDesc::UINT8);
465
466         out->open(filename, spec);
467
468         /* conversion for different top/bottom convention */
469         out->write_image(TypeDesc::UINT8,
470                 (uchar*)(pixels + (h-1)*w),
471                 AutoStride,
472                 -w*sizeof(uchar4),
473                 AutoStride);
474
475         out->close();
476
477         delete out;
478 }
479
480 CCL_NAMESPACE_END
481