Split out some private utilities into effect_util.cpp, so we do not need to include...
[movit] / ycbcr_input.cpp
1 #include <Eigen/Core>
2 #include <Eigen/LU>
3 #include <GL/glew.h>
4 #include <assert.h>
5 #include <stdio.h>
6 #include <string.h>
7
8 #include "effect_util.h"
9 #include "util.h"
10 #include "ycbcr_input.h"
11
12 using namespace Eigen;
13
14 namespace {
15
16 // OpenGL has texel center in (0.5, 0.5), but different formats have
17 // chroma in various other places. If luma samples are X, the chroma
18 // sample is *, and subsampling is 3x3, the situation with chroma
19 // center in (0.5, 0.5) looks approximately like this:
20 //
21 //   X   X
22 //     *   
23 //   X   X
24 //
25 // If, on the other hand, chroma center is in (0.0, 0.5) (common
26 // for e.g. MPEG-4), the figure changes to:
27 //
28 //   X   X
29 //   *      
30 //   X   X
31 //
32 // In other words, (0.0, 0.0) means that the chroma sample is exactly
33 // co-sited on top of the top-left luma sample. Note, however, that
34 // this is _not_ 0.5 texels to the left, since the OpenGL's texel center
35 // is in (0.5, 0.5); it is in (0.25, 0.25). In a sense, the four luma samples
36 // define a square where chroma position (0.0, 0.0) is in texel position
37 // (0.25, 0.25) and chroma position (1.0, 1.0) is in texel position (0.75, 0.75)
38 // (the outer border shows the borders of the texel itself, ie. from
39 // (0, 0) to (1, 1)):
40 //
41 //  ---------
42 // |         |
43 // |  X---X  |
44 // |  | * |  |
45 // |  X---X  |
46 // |         |
47 //  ---------
48 //
49 // Also note that if we have no subsampling, the square will have zero
50 // area and the chroma position does not matter at all.
51 float compute_chroma_offset(float pos, unsigned subsampling_factor, unsigned resolution)
52 {
53         float local_chroma_pos = (0.5 + pos * (subsampling_factor - 1)) / subsampling_factor;
54         return (0.5 - local_chroma_pos) / resolution;
55 }
56
57 }  // namespace
58
59 YCbCrInput::YCbCrInput(const ImageFormat &image_format,
60                        const YCbCrFormat &ycbcr_format,
61                        unsigned width, unsigned height)
62         : image_format(image_format),
63           ycbcr_format(ycbcr_format),
64           needs_update(false),
65           needs_pbo_recreate(false),
66           finalized(false),
67           needs_mipmaps(false),
68           width(width),
69           height(height)
70 {
71         pbos[0] = pbos[1] = pbos[2] = 0;
72         texture_num[0] = texture_num[1] = texture_num[2] = 0;
73
74         assert(width % ycbcr_format.chroma_subsampling_x == 0);
75         pitch[0] = widths[0] = width;
76         pitch[1] = widths[1] = width / ycbcr_format.chroma_subsampling_x;
77         pitch[2] = widths[2] = width / ycbcr_format.chroma_subsampling_x;
78
79         assert(height % ycbcr_format.chroma_subsampling_y == 0);
80         heights[0] = height;
81         heights[1] = height / ycbcr_format.chroma_subsampling_y;
82         heights[2] = height / ycbcr_format.chroma_subsampling_y;
83
84         pixel_data[0] = pixel_data[1] = pixel_data[2] = NULL;
85
86         register_int("needs_mipmaps", &needs_mipmaps);
87 }
88
89 YCbCrInput::~YCbCrInput()
90 {
91         if (pbos[0] != 0) {
92                 glDeleteBuffers(3, pbos);
93                 check_error();
94         }
95         if (texture_num[0] != 0) {
96                 glDeleteTextures(3, texture_num);
97                 check_error();
98         }
99 }
100
101 void YCbCrInput::finalize()
102 {
103         glPixelStorei(GL_UNPACK_ALIGNMENT, 1);
104         check_error();
105
106         // Create PBOs to hold the textures holding the input image, and then the texture itself.
107         glGenBuffers(3, pbos);
108         check_error();
109         glGenTextures(3, texture_num);
110         check_error();
111
112         for (unsigned channel = 0; channel < 3; ++channel) {
113                 glBindBuffer(GL_PIXEL_UNPACK_BUFFER_ARB, pbos[channel]);
114                 check_error();
115                 glBufferData(GL_PIXEL_UNPACK_BUFFER_ARB, pitch[channel] * heights[channel], NULL, GL_STREAM_DRAW);
116                 check_error();
117                 glBindBuffer(GL_PIXEL_UNPACK_BUFFER_ARB, 0);
118                 check_error();
119                 
120                 glBindTexture(GL_TEXTURE_2D, texture_num[channel]);
121                 check_error();
122                 glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
123                 check_error();
124                 glPixelStorei(GL_UNPACK_ROW_LENGTH, pitch[channel]);
125                 check_error();
126                 glTexImage2D(GL_TEXTURE_2D, 0, GL_LUMINANCE8, widths[channel], heights[channel], 0, GL_LUMINANCE, GL_UNSIGNED_BYTE, NULL);
127                 check_error();
128                 glPixelStorei(GL_UNPACK_ROW_LENGTH, 0);
129                 check_error();
130         }
131
132         needs_update = true;
133         finalized = true;
134 }
135         
136 void YCbCrInput::set_gl_state(GLuint glsl_program_num, const std::string& prefix, unsigned *sampler_num)
137 {
138         for (unsigned channel = 0; channel < 3; ++channel) {
139                 glActiveTexture(GL_TEXTURE0 + *sampler_num + channel);
140                 check_error();
141                 glBindTexture(GL_TEXTURE_2D, texture_num[channel]);
142                 check_error();
143
144                 if (needs_update || needs_pbo_recreate) {
145                         // Copy the pixel data into the PBO.
146                         glBindBuffer(GL_PIXEL_UNPACK_BUFFER_ARB, pbos[channel]);
147                         check_error();
148
149                         if (needs_pbo_recreate) {
150                                 // The pitch has changed; we need to reallocate this PBO.
151                                 glBufferData(GL_PIXEL_UNPACK_BUFFER_ARB, pitch[channel] * heights[channel], NULL, GL_STREAM_DRAW);
152                                 check_error();
153                         }
154
155                         void *mapped_pbo = glMapBufferARB(GL_PIXEL_UNPACK_BUFFER_ARB, GL_WRITE_ONLY);
156                         memcpy(mapped_pbo, pixel_data[channel], pitch[channel] * heights[channel]);
157
158                         glUnmapBufferARB(GL_PIXEL_UNPACK_BUFFER_ARB);
159                         check_error();
160
161                         // Re-upload the texture from the PBO.
162                         glPixelStorei(GL_UNPACK_ROW_LENGTH, pitch[channel]);
163                         check_error();
164                         glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, widths[channel], heights[channel], GL_LUMINANCE, GL_UNSIGNED_BYTE, BUFFER_OFFSET(0));
165                         check_error();
166                         glPixelStorei(GL_UNPACK_ROW_LENGTH, 0);
167                         check_error();
168                         glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
169                         check_error();
170                         glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
171                         check_error();
172                         glBindBuffer(GL_PIXEL_UNPACK_BUFFER_ARB, 0);
173                         check_error();
174                 }
175         }
176
177         // Bind samplers.
178         set_uniform_int(glsl_program_num, prefix, "tex_y", *sampler_num + 0);
179         set_uniform_int(glsl_program_num, prefix, "tex_cb", *sampler_num + 1);
180         set_uniform_int(glsl_program_num, prefix, "tex_cr", *sampler_num + 2);
181
182         *sampler_num += 3;
183         needs_update = false;
184         needs_pbo_recreate = false;
185 }
186
187 std::string YCbCrInput::output_fragment_shader()
188 {
189         float coeff[3], offset[3], scale[3];
190
191         switch (ycbcr_format.luma_coefficients) {
192         case YCBCR_REC_601:
193                 // Rec. 601, page 2.
194                 coeff[0] = 0.299;
195                 coeff[1] = 0.587;
196                 coeff[2] = 0.114;
197                 break;
198
199         case YCBCR_REC_709:
200                 // Rec. 709, page 19.
201                 coeff[0] = 0.2126;
202                 coeff[1] = 0.7152;
203                 coeff[2] = 0.0722;
204                 break;
205         default:
206                 assert(false);
207         }
208
209         if (ycbcr_format.full_range) {
210                 offset[0] = 0.0 / 255.0;
211                 offset[1] = 128.0 / 255.0;
212                 offset[2] = 128.0 / 255.0;
213
214                 scale[0] = 1.0;
215                 scale[1] = 1.0;
216                 scale[2] = 1.0;
217         } else {
218                 // Rec. 601, page 4; Rec. 709, page 19.
219                 offset[0] = 16.0 / 255.0;
220                 offset[1] = 128.0 / 255.0;
221                 offset[2] = 128.0 / 255.0;
222
223                 scale[0] = 255.0 / 219.0;
224                 scale[1] = 255.0 / 224.0;
225                 scale[2] = 255.0 / 224.0;
226         }
227
228         // Matrix to convert RGB to YCbCr. See e.g. Rec. 601.
229         Matrix3d rgb_to_ycbcr;
230         rgb_to_ycbcr(0,0) = coeff[0];
231         rgb_to_ycbcr(0,1) = coeff[1];
232         rgb_to_ycbcr(0,2) = coeff[2];
233
234         float cb_fac = (224.0 / 219.0) / (coeff[0] + coeff[1] + 1.0f - coeff[2]);
235         rgb_to_ycbcr(1,0) = -coeff[0] * cb_fac;
236         rgb_to_ycbcr(1,1) = -coeff[1] * cb_fac;
237         rgb_to_ycbcr(1,2) = (1.0f - coeff[2]) * cb_fac;
238
239         float cr_fac = (224.0 / 219.0) / (1.0f - coeff[0] + coeff[1] + coeff[2]);
240         rgb_to_ycbcr(2,0) = (1.0f - coeff[0]) * cr_fac;
241         rgb_to_ycbcr(2,1) = -coeff[1] * cr_fac;
242         rgb_to_ycbcr(2,2) = -coeff[2] * cr_fac;
243
244         // Inverting the matrix gives us what we need to go from YCbCr back to RGB.
245         Matrix3d ycbcr_to_rgb = rgb_to_ycbcr.inverse();
246
247         std::string frag_shader;
248
249         frag_shader = output_glsl_mat3("PREFIX(inv_ycbcr_matrix)", ycbcr_to_rgb);
250
251         char buf[256];
252         sprintf(buf, "const vec3 PREFIX(offset) = vec3(%.8f, %.8f, %.8f);\n",
253                 offset[0], offset[1], offset[2]);
254         frag_shader += buf;
255
256         sprintf(buf, "const vec3 PREFIX(scale) = vec3(%.8f, %.8f, %.8f);\n",
257                 scale[0], scale[1], scale[2]);
258         frag_shader += buf;
259
260         float cb_offset_x = compute_chroma_offset(
261                 ycbcr_format.cb_x_position, ycbcr_format.chroma_subsampling_x, widths[1]);
262         float cb_offset_y = compute_chroma_offset(
263                 ycbcr_format.cb_y_position, ycbcr_format.chroma_subsampling_y, heights[1]);
264         sprintf(buf, "const vec2 PREFIX(cb_offset) = vec2(%.8f, %.8f);\n",
265                 cb_offset_x, cb_offset_y);
266         frag_shader += buf;
267
268         float cr_offset_x = compute_chroma_offset(
269                 ycbcr_format.cr_x_position, ycbcr_format.chroma_subsampling_x, widths[2]);
270         float cr_offset_y = compute_chroma_offset(
271                 ycbcr_format.cr_y_position, ycbcr_format.chroma_subsampling_y, heights[2]);
272         sprintf(buf, "const vec2 PREFIX(cr_offset) = vec2(%.8f, %.8f);\n",
273                 cr_offset_x, cr_offset_y);
274         frag_shader += buf;
275
276         frag_shader += read_file("ycbcr_input.frag");
277         return frag_shader;
278 }