X-Git-Url: https://git.sesse.net/?p=movit;a=blobdiff_plain;f=util.cpp;h=59f1dcdd67529ef9b4fd54f959fdc49bdc41d633;hp=ac6a218098873953d25036605d728ed98257ec70;hb=6f1efa8348a90a393187c12d70fd10d81bbd2c99;hpb=144496bf79000c3971090195fbfedcb2cb22a0be diff --git a/util.cpp b/util.cpp index ac6a218..59f1dcd 100644 --- a/util.cpp +++ b/util.cpp @@ -4,11 +4,23 @@ #include #include #include +#include +#include +#include #include +#include "fp16.h" #include "init.h" #include "util.h" +#if defined(__APPLE__) +#include +#elif defined(WIN32) +#include +#else +#include +#endif + using namespace std; namespace movit { @@ -75,17 +87,54 @@ 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) { + return read_file(base + ".130." + extension); + } else if (movit_shader_model == MOVIT_ESSL_300) { + return read_file(base + ".300es." + extension); + } else { + assert(false); + } } GLuint compile_shader(const string &shader_src, GLenum type) @@ -107,6 +156,7 @@ 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); } @@ -123,34 +173,72 @@ 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_fp64(pos1 + z * (pos2 - pos1)); + z = (to_fp64(*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: @@ -165,19 +253,24 @@ 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_fp64((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_fp64(*total_weight) * (1 - z) - w1; + float err2 = to_fp64(*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 fill_vertex_attribute(GLuint glsl_program_num, const string &attribute_name, GLint size, GLenum type, GLsizeiptr data_size, const GLvoid *data) { int attrib = glGetAttribLocation(glsl_program_num, attribute_name.c_str()); @@ -233,4 +326,15 @@ unsigned next_power_of_two(unsigned v) return v; } +void *get_gl_context_identifier() +{ +#if defined(__APPLE__) + return (void *)CGLGetCurrentContext(); +#elif defined(WIN32) + return (void *)wglGetCurrentContext(); +#else + return (void *)glXGetCurrentContext(); +#endif +} + } // namespace movit