X-Git-Url: https://git.sesse.net/?p=movit;a=blobdiff_plain;f=util.cpp;h=3141f8d2c5eb388aec128f31b6218c06c0b1bb8d;hp=01787f013e0a4c70817be2ccf49743cddb7bd254;hb=refs%2Fheads%2F1.3.x-release;hpb=ddf71f853e64c3912eed4ab98bfe7503826ce8e1 diff --git a/util.cpp b/util.cpp index 01787f0..3141f8d 100644 --- a/util.cpp +++ b/util.cpp @@ -4,17 +4,22 @@ #include #include #include +#include +#include +#include #include +#include "fp16.h" #include "init.h" #include "util.h" -#if defined(__DARWIN__) +#if defined(__APPLE__) #include #elif defined(WIN32) #include #else #include +#include #endif using namespace std; @@ -83,25 +88,49 @@ string read_file(const string &filename) { const string full_pathname = *movit_data_directory + "/" + filename; - static char buf[131072]; FILE *fp = fopen(full_pathname.c_str(), "r"); if (fp == NULL) { perror(full_pathname.c_str()); exit(1); } - int len = fread(buf, 1, sizeof(buf), fp); + int ret = fseek(fp, 0, SEEK_END); + if (ret == -1) { + perror("fseek(SEEK_END)"); + exit(1); + } + + int size = ftell(fp); + + ret = fseek(fp, 0, SEEK_SET); + if (ret == -1) { + perror("fseek(SEEK_SET)"); + exit(1); + } + + string str; + str.resize(size); + ret = fread(&str[0], size, 1, fp); + if (ret == -1) { + perror("fread"); + exit(1); + } + if (ret == 0) { + fprintf(stderr, "Short read when trying to read %d bytes from %s\n", + size, full_pathname.c_str()); + exit(1); + } fclose(fp); - return string(buf, len); + return str; } string read_version_dependent_file(const string &base, const string &extension) { - if (movit_shader_model == MOVIT_GLSL_110) { - return read_file(base + "." + extension); - } else if (movit_shader_model == MOVIT_GLSL_130) { + if (movit_shader_model == MOVIT_GLSL_130) { return read_file(base + ".130." + extension); + } else if (movit_shader_model == MOVIT_GLSL_150) { + return read_file(base + ".150." + extension); } else if (movit_shader_model == MOVIT_ESSL_300) { return read_file(base + ".300es." + extension); } else { @@ -128,6 +157,7 @@ GLuint compile_shader(const string &shader_src, GLenum type) GLint status; glGetShaderiv(obj, GL_COMPILE_STATUS, &status); if (status == GL_FALSE) { + fprintf(stderr, "Failed to compile shader: %s\n", shader_src.c_str()); exit(1); } @@ -144,34 +174,72 @@ void print_3x3_matrix(const Eigen::Matrix3d& m) string output_glsl_mat3(const string &name, const Eigen::Matrix3d &m) { - char buf[1024]; - sprintf(buf, - "const mat3 %s = mat3(\n" - " %.8f, %.8f, %.8f,\n" - " %.8f, %.8f, %.8f,\n" - " %.8f, %.8f, %.8f);\n\n", - name.c_str(), - m(0,0), m(1,0), m(2,0), - m(0,1), m(1,1), m(2,1), - m(0,2), m(1,2), m(2,2)); - return buf; + // Use stringstream to be independent of the current locale in a thread-safe manner. + stringstream ss; + ss.imbue(locale("C")); + ss.precision(8); + ss << scientific; + ss << "const mat3 " << name << " = mat3(\n"; + ss << " " << m(0,0) << ", " << m(1,0) << ", " << m(2,0) << ",\n"; + ss << " " << m(0,1) << ", " << m(1,1) << ", " << m(2,1) << ",\n"; + ss << " " << m(0,2) << ", " << m(1,2) << ", " << m(2,2) << ");\n\n"; + return ss.str(); } -void combine_two_samples(float w1, float w2, float *offset, float *total_weight, float *sum_sq_error) +string output_glsl_float(const string &name, float x) +{ + // Use stringstream to be independent of the current locale in a thread-safe manner. + stringstream ss; + ss.imbue(locale("C")); + ss.precision(8); + ss << scientific; + ss << "const float " << name << " = " << x << ";\n"; + return ss.str(); +} + +string output_glsl_vec2(const string &name, float x, float y) +{ + // Use stringstream to be independent of the current locale in a thread-safe manner. + stringstream ss; + ss.imbue(locale("C")); + ss.precision(8); + ss << scientific; + ss << "const vec2 " << name << " = vec2(" << x << ", " << y << ");\n"; + return ss.str(); +} + +string output_glsl_vec3(const string &name, float x, float y, float z) +{ + // Use stringstream to be independent of the current locale in a thread-safe manner. + stringstream ss; + ss.imbue(locale("C")); + ss.precision(8); + ss << scientific; + ss << "const vec3 " << name << " = vec3(" << x << ", " << y << ", " << z << ");\n"; + return ss.str(); +} + +template +void combine_two_samples(float w1, float w2, float pos1, float pos2, float num_subtexels, float inv_num_subtexels, + DestFloat *offset, DestFloat *total_weight, float *sum_sq_error) { assert(movit_initialized); assert(w1 * w2 >= 0.0f); // Should not have differing signs. - float z; // Just a shorter name for offset. + float z; // Normalized 0..1 between pos1 and pos2. if (fabs(w1 + w2) < 1e-6) { z = 0.5f; } else { z = w2 / (w1 + w2); } + // Round to the desired precision. Note that this might take z outside the 0..1 range. + *offset = from_fp32(pos1 + z * (pos2 - pos1)); + z = (to_fp32(*offset) - pos1) / (pos2 - pos1); + // Round to the minimum number of bits we have measured earlier. // The card will do this for us anyway, but if we know what the real z // is, we can pick a better total_weight below. - z = lrintf(z / movit_texel_subpixel_precision) * movit_texel_subpixel_precision; + z = lrintf(z * num_subtexels) * inv_num_subtexels; // Choose total weight w so that we minimize total squared error // for the effective weights: @@ -186,17 +254,37 @@ void combine_two_samples(float w1, float w2, float *offset, float *total_weight, // w = (a(1-z) + bz) / ((1-z)² + z²) // // If z had infinite precision, this would simply reduce to w = w1 + w2. - *total_weight = (w1 * (1 - z) + w2 * z) / (z * z + (1 - z) * (1 - z)); - *offset = z; + *total_weight = from_fp32((w1 + z * (w2 - w1)) / (z * z + (1 - z) * (1 - z))); if (sum_sq_error != NULL) { - float err1 = *total_weight * (1 - z) - w1; - float err2 = *total_weight * z - w2; + float err1 = to_fp32(*total_weight) * (1 - z) - w1; + float err2 = to_fp32(*total_weight) * z - w2; *sum_sq_error = err1 * err1 + err2 * err2; } +} - assert(*offset >= 0.0f); - assert(*offset <= 1.0f); +// Explicit instantiations. +template +void combine_two_samples(float w1, float w2, float pos1, float pos2, float num_subtexels, float inv_num_subtexels, + float *offset, float *total_weight, float *sum_sq_error); + +template +void combine_two_samples(float w1, float w2, float pos1, float pos2, float num_subtexels, float inv_num_subtexels, + fp16_int_t *offset, fp16_int_t *total_weight, float *sum_sq_error); + +GLuint generate_vbo(GLint size, GLenum type, GLsizeiptr data_size, const GLvoid *data) +{ + GLuint vbo; + glGenBuffers(1, &vbo); + check_error(); + glBindBuffer(GL_ARRAY_BUFFER, vbo); + check_error(); + glBufferData(GL_ARRAY_BUFFER, data_size, data, GL_STATIC_DRAW); + check_error(); + glBindBuffer(GL_ARRAY_BUFFER, 0); + check_error(); + + return vbo; } GLuint fill_vertex_attribute(GLuint glsl_program_num, const string &attribute_name, GLint size, GLenum type, GLsizeiptr data_size, const GLvoid *data) @@ -206,13 +294,10 @@ GLuint fill_vertex_attribute(GLuint glsl_program_num, const string &attribute_na return -1; } - GLuint vbo; - glGenBuffers(1, &vbo); - check_error(); + GLuint vbo = generate_vbo(size, type, data_size, data); + glBindBuffer(GL_ARRAY_BUFFER, vbo); check_error(); - glBufferData(GL_ARRAY_BUFFER, data_size, data, GL_STATIC_DRAW); - check_error(); glEnableVertexAttribArray(attrib); check_error(); glVertexAttribPointer(attrib, size, type, GL_FALSE, 0, BUFFER_OFFSET(0)); @@ -256,11 +341,15 @@ unsigned next_power_of_two(unsigned v) void *get_gl_context_identifier() { -#if defined(__DARWIN__) +#if defined(__APPLE__) return (void *)CGLGetCurrentContext(); #elif defined(WIN32) return (void *)wglGetCurrentContext(); #else + void *ret = (void *)eglGetCurrentContext(); + if (ret != NULL) { + return ret; + } return (void *)glXGetCurrentContext(); #endif }