16 #if defined(__APPLE__)
17 #include <OpenGL/OpenGL.h>
19 #include <epoxy/wgl.h>
21 #include <epoxy/glx.h>
22 #include <epoxy/egl.h>
29 extern string *movit_data_directory;
31 void hsv2rgb(float h, float s, float v, float *r, float *g, float *b)
37 float hp = (h * 180.0 / M_PI) / 60.0;
38 float x = c * (1 - fabs(fmod(hp, 2.0f) - 1.0f));
40 if (hp >= 0 && hp < 1) {
44 } else if (hp >= 1 && hp < 2) {
48 } else if (hp >= 2 && hp < 3) {
52 } else if (hp >= 3 && hp < 4) {
56 } else if (hp >= 4 && hp < 5) {
72 void hsv2rgb_normalized(float h, float s, float v, float *r, float *g, float *b)
74 float ref_r, ref_g, ref_b;
75 hsv2rgb(h, s, v, r, g, b);
76 hsv2rgb(h, 0.0f, v, &ref_r, &ref_g, &ref_b);
77 float lum = 0.2126 * *r + 0.7152 * *g + 0.0722 * *b;
78 float ref_lum = 0.2126 * ref_r + 0.7152 * ref_g + 0.0722 * ref_b;
80 float fac = ref_lum / lum;
87 string read_file(const string &filename)
89 const string full_pathname = *movit_data_directory + "/" + filename;
91 FILE *fp = fopen(full_pathname.c_str(), "r");
93 perror(full_pathname.c_str());
97 int ret = fseek(fp, 0, SEEK_END);
99 perror("fseek(SEEK_END)");
103 int size = ftell(fp);
105 ret = fseek(fp, 0, SEEK_SET);
107 perror("fseek(SEEK_SET)");
113 ret = fread(&str[0], size, 1, fp);
119 fprintf(stderr, "Short read when trying to read %d bytes from %s\n",
120 size, full_pathname.c_str());
128 string read_version_dependent_file(const string &base, const string &extension)
130 if (movit_shader_model == MOVIT_GLSL_130) {
131 return read_file(base + ".130." + extension);
132 } else if (movit_shader_model == MOVIT_GLSL_150) {
133 return read_file(base + ".150." + extension);
134 } else if (movit_shader_model == MOVIT_ESSL_300) {
135 return read_file(base + ".300es." + extension);
141 GLuint compile_shader(const string &shader_src, GLenum type)
143 GLuint obj = glCreateShader(type);
144 const GLchar* source[] = { shader_src.data() };
145 const GLint length[] = { (GLint)shader_src.size() };
146 glShaderSource(obj, 1, source, length);
147 glCompileShader(obj);
149 GLchar info_log[4096];
150 GLsizei log_length = sizeof(info_log) - 1;
151 glGetShaderInfoLog(obj, log_length, &log_length, info_log);
152 info_log[log_length] = 0;
153 if (strlen(info_log) > 0) {
154 fprintf(stderr, "Shader compile log: %s\n", info_log);
158 glGetShaderiv(obj, GL_COMPILE_STATUS, &status);
159 if (status == GL_FALSE) {
160 fprintf(stderr, "Failed to compile shader: %s\n", shader_src.c_str());
167 void print_3x3_matrix(const Eigen::Matrix3d& m)
169 printf("%6.4f %6.4f %6.4f\n", m(0,0), m(0,1), m(0,2));
170 printf("%6.4f %6.4f %6.4f\n", m(1,0), m(1,1), m(1,2));
171 printf("%6.4f %6.4f %6.4f\n", m(2,0), m(2,1), m(2,2));
175 string output_glsl_mat3(const string &name, const Eigen::Matrix3d &m)
177 // Use stringstream to be independent of the current locale in a thread-safe manner.
179 ss.imbue(locale("C"));
182 ss << "const mat3 " << name << " = mat3(\n";
183 ss << " " << m(0,0) << ", " << m(1,0) << ", " << m(2,0) << ",\n";
184 ss << " " << m(0,1) << ", " << m(1,1) << ", " << m(2,1) << ",\n";
185 ss << " " << m(0,2) << ", " << m(1,2) << ", " << m(2,2) << ");\n\n";
189 string output_glsl_float(const string &name, float x)
191 // Use stringstream to be independent of the current locale in a thread-safe manner.
193 ss.imbue(locale("C"));
196 ss << "const float " << name << " = " << x << ";\n";
200 string output_glsl_vec2(const string &name, float x, float y)
202 // Use stringstream to be independent of the current locale in a thread-safe manner.
204 ss.imbue(locale("C"));
207 ss << "const vec2 " << name << " = vec2(" << x << ", " << y << ");\n";
211 string output_glsl_vec3(const string &name, float x, float y, float z)
213 // Use stringstream to be independent of the current locale in a thread-safe manner.
215 ss.imbue(locale("C"));
218 ss << "const vec3 " << name << " = vec3(" << x << ", " << y << ", " << z << ");\n";
222 template<class DestFloat>
223 void combine_two_samples(float w1, float w2, float pos1, float pos2, float num_subtexels, float inv_num_subtexels,
224 DestFloat *offset, DestFloat *total_weight, float *sum_sq_error)
226 assert(movit_initialized);
227 assert(w1 * w2 >= 0.0f); // Should not have differing signs.
228 float z; // Normalized 0..1 between pos1 and pos2.
229 if (fabs(w1 + w2) < 1e-6) {
235 // Round to the desired precision. Note that this might take z outside the 0..1 range.
236 *offset = from_fp32<DestFloat>(pos1 + z * (pos2 - pos1));
237 z = (to_fp32(*offset) - pos1) / (pos2 - pos1);
239 // Round to the minimum number of bits we have measured earlier.
240 // The card will do this for us anyway, but if we know what the real z
241 // is, we can pick a better total_weight below.
242 z = lrintf(z * num_subtexels) * inv_num_subtexels;
244 // Choose total weight w so that we minimize total squared error
245 // for the effective weights:
247 // e = (w(1-z) - a)² + (wz - b)²
249 // Differentiating by w and setting equal to zero:
251 // 2(w(1-z) - a)(1-z) + 2(wz - b)z = 0
252 // w(1-z)² - a(1-z) + wz² - bz = 0
253 // w((1-z)² + z²) = a(1-z) + bz
254 // w = (a(1-z) + bz) / ((1-z)² + z²)
256 // If z had infinite precision, this would simply reduce to w = w1 + w2.
257 *total_weight = from_fp32<DestFloat>((w1 + z * (w2 - w1)) / (z * z + (1 - z) * (1 - z)));
259 if (sum_sq_error != NULL) {
260 float err1 = to_fp32(*total_weight) * (1 - z) - w1;
261 float err2 = to_fp32(*total_weight) * z - w2;
262 *sum_sq_error = err1 * err1 + err2 * err2;
266 // Explicit instantiations.
268 void combine_two_samples<float>(float w1, float w2, float pos1, float pos2, float num_subtexels, float inv_num_subtexels,
269 float *offset, float *total_weight, float *sum_sq_error);
272 void combine_two_samples<fp16_int_t>(float w1, float w2, float pos1, float pos2, float num_subtexels, float inv_num_subtexels,
273 fp16_int_t *offset, fp16_int_t *total_weight, float *sum_sq_error);
275 GLuint generate_vbo(GLint size, GLenum type, GLsizeiptr data_size, const GLvoid *data)
278 glGenBuffers(1, &vbo);
280 glBindBuffer(GL_ARRAY_BUFFER, vbo);
282 glBufferData(GL_ARRAY_BUFFER, data_size, data, GL_STATIC_DRAW);
284 glBindBuffer(GL_ARRAY_BUFFER, 0);
290 GLuint fill_vertex_attribute(GLuint glsl_program_num, const string &attribute_name, GLint size, GLenum type, GLsizeiptr data_size, const GLvoid *data)
292 int attrib = glGetAttribLocation(glsl_program_num, attribute_name.c_str());
297 GLuint vbo = generate_vbo(size, type, data_size, data);
299 glBindBuffer(GL_ARRAY_BUFFER, vbo);
301 glEnableVertexAttribArray(attrib);
303 glVertexAttribPointer(attrib, size, type, GL_FALSE, 0, BUFFER_OFFSET(0));
305 glBindBuffer(GL_ARRAY_BUFFER, 0);
311 void cleanup_vertex_attribute(GLuint glsl_program_num, const string &attribute_name, GLuint vbo)
313 int attrib = glGetAttribLocation(glsl_program_num, attribute_name.c_str());
318 glDisableVertexAttribArray(attrib);
320 glDeleteBuffers(1, &vbo);
324 unsigned div_round_up(unsigned a, unsigned b)
326 return (a + b - 1) / b;
329 // Algorithm from http://graphics.stanford.edu/~seander/bithacks.html#RoundUpPowerOf2.
330 unsigned next_power_of_two(unsigned v)
342 void *get_gl_context_identifier()
344 #if defined(__APPLE__)
345 return (void *)CGLGetCurrentContext();
347 return (void *)wglGetCurrentContext();
349 void *ret = (void *)eglGetCurrentContext();
353 return (void *)glXGetCurrentContext();