#include <stdio.h>
#include <stdlib.h>
#include <string.h>
+#include <locale>
+#include <sstream>
+#include <string>
#include <Eigen/Core>
#include "fp16.h"
{
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)
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);
}
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();
+}
+
+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();
}
-void combine_two_samples(float w1, float w2, float pos1, float pos2, unsigned size, CombineRoundingBehavior rounding_behavior,
- float *offset, float *total_weight, float *sum_sq_error)
+template<class DestFloat>
+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.
z = w2 / (w1 + w2);
}
- *offset = pos1 + z * (pos2 - pos1);
- if (rounding_behavior == COMBINE_ROUND_TO_FP16) {
- // Round to fp16. Note that this might take z outside the 0..1 range.
- *offset = fp16_to_fp64(fp64_to_fp16(*offset));
- z = (*offset - pos1) / (pos2 - pos1);
- } else {
- assert(rounding_behavior == COMBINE_DO_NOT_ROUND);
- }
+ // Round to the desired precision. Note that this might take z outside the 0..1 range.
+ *offset = from_fp64<DestFloat>(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 *= size; // Move to pixel coordinates,
- z = lrintf(z / movit_texel_subpixel_precision) * movit_texel_subpixel_precision; // Round.
- z /= size; // Move back to normalized coordinates.
+ z = lrintf(z * num_subtexels) * inv_num_subtexels;
// Choose total weight w so that we minimize total squared error
// for the effective weights:
// 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));
+ *total_weight = from_fp64<DestFloat>((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;
}
}
+// Explicit instantiations.
+template
+void combine_two_samples<float>(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<fp16_int_t>(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());
projects / movit / blobdiff