Implement the texture freelist in ResourcePool.
[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 "resource_pool.h"
10 #include "util.h"
11 #include "ycbcr_input.h"
12
13 using namespace Eigen;
14
15 namespace {
16
17 // OpenGL has texel center in (0.5, 0.5), but different formats have
18 // chroma in various other places. If luma samples are X, the chroma
19 // sample is *, and subsampling is 3x3, the situation with chroma
20 // center in (0.5, 0.5) looks approximately like this:
21 //
22 //   X   X
23 //     *   
24 //   X   X
25 //
26 // If, on the other hand, chroma center is in (0.0, 0.5) (common
27 // for e.g. MPEG-4), the figure changes to:
28 //
29 //   X   X
30 //   *      
31 //   X   X
32 //
33 // In other words, (0.0, 0.0) means that the chroma sample is exactly
34 // co-sited on top of the top-left luma sample. Note, however, that
35 // this is _not_ 0.5 texels to the left, since the OpenGL's texel center
36 // is in (0.5, 0.5); it is in (0.25, 0.25). In a sense, the four luma samples
37 // define a square where chroma position (0.0, 0.0) is in texel position
38 // (0.25, 0.25) and chroma position (1.0, 1.0) is in texel position (0.75, 0.75)
39 // (the outer border shows the borders of the texel itself, ie. from
40 // (0, 0) to (1, 1)):
41 //
42 //  ---------
43 // |         |
44 // |  X---X  |
45 // |  | * |  |
46 // |  X---X  |
47 // |         |
48 //  ---------
49 //
50 // Also note that if we have no subsampling, the square will have zero
51 // area and the chroma position does not matter at all.
52 float compute_chroma_offset(float pos, unsigned subsampling_factor, unsigned resolution)
53 {
54         float local_chroma_pos = (0.5 + pos * (subsampling_factor - 1)) / subsampling_factor;
55         return (0.5 - local_chroma_pos) / resolution;
56 }
57
58 }  // namespace
59
60 YCbCrInput::YCbCrInput(const ImageFormat &image_format,
61                        const YCbCrFormat &ycbcr_format,
62                        unsigned width, unsigned height)
63         : image_format(image_format),
64           ycbcr_format(ycbcr_format),
65           needs_update(false),
66           finalized(false),
67           needs_mipmaps(false),
68           width(width),
69           height(height),
70           resource_pool(NULL)
71 {
72         pbos[0] = pbos[1] = pbos[2] = 0;
73         texture_num[0] = texture_num[1] = texture_num[2] = 0;
74
75         assert(width % ycbcr_format.chroma_subsampling_x == 0);
76         pitch[0] = widths[0] = width;
77         pitch[1] = widths[1] = width / ycbcr_format.chroma_subsampling_x;
78         pitch[2] = widths[2] = width / ycbcr_format.chroma_subsampling_x;
79
80         assert(height % ycbcr_format.chroma_subsampling_y == 0);
81         heights[0] = height;
82         heights[1] = height / ycbcr_format.chroma_subsampling_y;
83         heights[2] = height / ycbcr_format.chroma_subsampling_y;
84
85         pixel_data[0] = pixel_data[1] = pixel_data[2] = NULL;
86
87         register_int("needs_mipmaps", &needs_mipmaps);
88 }
89
90 YCbCrInput::~YCbCrInput()
91 {
92         for (unsigned channel = 0; channel < 3; ++channel) {
93                 if (texture_num[channel] != 0) {
94                         resource_pool->release_2d_texture(texture_num[channel]);
95                 }
96         }
97 }
98
99 void YCbCrInput::finalize()
100 {
101         // Create the textures themselves.
102         for (unsigned channel = 0; channel < 3; ++channel) {
103                 texture_num[channel] = resource_pool->create_2d_texture(GL_LUMINANCE8, widths[channel], heights[channel]);
104                 glBindTexture(GL_TEXTURE_2D, texture_num[channel]);
105                 check_error();
106                 glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
107                 check_error();
108         }
109
110         needs_update = true;
111         finalized = true;
112 }
113         
114 void YCbCrInput::set_gl_state(GLuint glsl_program_num, const std::string& prefix, unsigned *sampler_num)
115 {
116         for (unsigned channel = 0; channel < 3; ++channel) {
117                 glActiveTexture(GL_TEXTURE0 + *sampler_num + channel);
118                 check_error();
119                 glBindTexture(GL_TEXTURE_2D, texture_num[channel]);
120                 check_error();
121
122                 if (needs_update) {
123                         // Re-upload the texture.
124                         // Copy the pixel data into the PBO.
125                         glBindBuffer(GL_PIXEL_UNPACK_BUFFER_ARB, pbos[channel]);
126                         check_error();
127                         glPixelStorei(GL_UNPACK_ALIGNMENT, 1);
128                         check_error();
129                         glPixelStorei(GL_UNPACK_ROW_LENGTH, pitch[channel]);
130                         check_error();
131                         glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, widths[channel], heights[channel], GL_LUMINANCE, GL_UNSIGNED_BYTE, pixel_data[channel]);
132                         check_error();
133                         glPixelStorei(GL_UNPACK_ROW_LENGTH, 0);
134                         check_error();
135                         glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
136                         check_error();
137                         glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
138                         check_error();
139                 }
140         }
141
142         glBindBuffer(GL_PIXEL_UNPACK_BUFFER_ARB, 0);
143         check_error();
144
145         // Bind samplers.
146         set_uniform_int(glsl_program_num, prefix, "tex_y", *sampler_num + 0);
147         set_uniform_int(glsl_program_num, prefix, "tex_cb", *sampler_num + 1);
148         set_uniform_int(glsl_program_num, prefix, "tex_cr", *sampler_num + 2);
149
150         *sampler_num += 3;
151         needs_update = false;
152 }
153
154 std::string YCbCrInput::output_fragment_shader()
155 {
156         float coeff[3], offset[3], scale[3];
157
158         switch (ycbcr_format.luma_coefficients) {
159         case YCBCR_REC_601:
160                 // Rec. 601, page 2.
161                 coeff[0] = 0.299;
162                 coeff[1] = 0.587;
163                 coeff[2] = 0.114;
164                 break;
165
166         case YCBCR_REC_709:
167                 // Rec. 709, page 19.
168                 coeff[0] = 0.2126;
169                 coeff[1] = 0.7152;
170                 coeff[2] = 0.0722;
171                 break;
172
173         case YCBCR_REC_2020:
174                 // Rec. 2020, page 4.
175                 coeff[0] = 0.2627;
176                 coeff[1] = 0.6780;
177                 coeff[2] = 0.0593;
178                 break;
179
180         default:
181                 assert(false);
182         }
183
184         if (ycbcr_format.full_range) {
185                 offset[0] = 0.0 / 255.0;
186                 offset[1] = 128.0 / 255.0;
187                 offset[2] = 128.0 / 255.0;
188
189                 scale[0] = 1.0;
190                 scale[1] = 1.0;
191                 scale[2] = 1.0;
192         } else {
193                 // Rec. 601, page 4; Rec. 709, page 19; Rec. 2020, page 4.
194                 offset[0] = 16.0 / 255.0;
195                 offset[1] = 128.0 / 255.0;
196                 offset[2] = 128.0 / 255.0;
197
198                 scale[0] = 255.0 / 219.0;
199                 scale[1] = 255.0 / 224.0;
200                 scale[2] = 255.0 / 224.0;
201         }
202
203         // Matrix to convert RGB to YCbCr. See e.g. Rec. 601.
204         Matrix3d rgb_to_ycbcr;
205         rgb_to_ycbcr(0,0) = coeff[0];
206         rgb_to_ycbcr(0,1) = coeff[1];
207         rgb_to_ycbcr(0,2) = coeff[2];
208
209         float cb_fac = (224.0 / 219.0) / (coeff[0] + coeff[1] + 1.0f - coeff[2]);
210         rgb_to_ycbcr(1,0) = -coeff[0] * cb_fac;
211         rgb_to_ycbcr(1,1) = -coeff[1] * cb_fac;
212         rgb_to_ycbcr(1,2) = (1.0f - coeff[2]) * cb_fac;
213
214         float cr_fac = (224.0 / 219.0) / (1.0f - coeff[0] + coeff[1] + coeff[2]);
215         rgb_to_ycbcr(2,0) = (1.0f - coeff[0]) * cr_fac;
216         rgb_to_ycbcr(2,1) = -coeff[1] * cr_fac;
217         rgb_to_ycbcr(2,2) = -coeff[2] * cr_fac;
218
219         // Inverting the matrix gives us what we need to go from YCbCr back to RGB.
220         Matrix3d ycbcr_to_rgb = rgb_to_ycbcr.inverse();
221
222         std::string frag_shader;
223
224         frag_shader = output_glsl_mat3("PREFIX(inv_ycbcr_matrix)", ycbcr_to_rgb);
225
226         char buf[256];
227         sprintf(buf, "const vec3 PREFIX(offset) = vec3(%.8f, %.8f, %.8f);\n",
228                 offset[0], offset[1], offset[2]);
229         frag_shader += buf;
230
231         sprintf(buf, "const vec3 PREFIX(scale) = vec3(%.8f, %.8f, %.8f);\n",
232                 scale[0], scale[1], scale[2]);
233         frag_shader += buf;
234
235         float cb_offset_x = compute_chroma_offset(
236                 ycbcr_format.cb_x_position, ycbcr_format.chroma_subsampling_x, widths[1]);
237         float cb_offset_y = compute_chroma_offset(
238                 ycbcr_format.cb_y_position, ycbcr_format.chroma_subsampling_y, heights[1]);
239         sprintf(buf, "const vec2 PREFIX(cb_offset) = vec2(%.8f, %.8f);\n",
240                 cb_offset_x, cb_offset_y);
241         frag_shader += buf;
242
243         float cr_offset_x = compute_chroma_offset(
244                 ycbcr_format.cr_x_position, ycbcr_format.chroma_subsampling_x, widths[2]);
245         float cr_offset_y = compute_chroma_offset(
246                 ycbcr_format.cr_y_position, ycbcr_format.chroma_subsampling_y, heights[2]);
247         sprintf(buf, "const vec2 PREFIX(cr_offset) = vec2(%.8f, %.8f);\n",
248                 cr_offset_x, cr_offset_y);
249         frag_shader += buf;
250
251         frag_shader += read_file("ycbcr_input.frag");
252         return frag_shader;
253 }