Convert a loop to range-based for.
[movit] / init.cpp
index 129ee4c..5e6c642 100644 (file)
--- a/init.cpp
+++ b/init.cpp
@@ -1,15 +1,35 @@
+#include <epoxy/gl.h>
+#include <assert.h>
+#include <stddef.h>
+#include <algorithm>
+#include <string>
+
 #include "init.h"
-#include "opengl.h"
+#include "resource_pool.h"
 #include "util.h"
 
+using namespace std;
+
+namespace movit {
+
 bool movit_initialized = false;
+MovitDebugLevel movit_debug_level = MOVIT_DEBUG_ON;
 float movit_texel_subpixel_precision;
+bool movit_timer_queries_supported, movit_compute_shaders_supported;
+int movit_num_wrongly_rounded;
+MovitShaderModel movit_shader_model;
+
+// The rules for objects with nontrivial constructors in static scope
+// are somewhat convoluted, and easy to mess up. We simply have a
+// pointer instead (and never care to clean it up).
+string *movit_data_directory = nullptr;
 
 namespace {
 
 void measure_texel_subpixel_precision()
 {
-       static const unsigned width = 1024;
+       ResourcePool resource_pool;
+       static const unsigned width = 4096;
 
        // Generate a destination texture to render to, and an FBO.
        GLuint dst_texnum, fbo;
@@ -18,7 +38,7 @@ void measure_texel_subpixel_precision()
        check_error();
        glBindTexture(GL_TEXTURE_2D, dst_texnum);
        check_error();
-       glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA16F_ARB, width, 1, 0, GL_RGBA, GL_UNSIGNED_BYTE, NULL);
+       glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA16F, width, 1, 0, GL_RGBA, GL_FLOAT, nullptr);
        check_error();
 
        glGenFramebuffers(1, &fbo);
@@ -36,17 +56,15 @@ void measure_texel_subpixel_precision()
        // Now generate a simple texture that's just [0,1].
        GLuint src_texnum;
        float texdata[] = { 0, 1 };
-       glGenTextures(1, &dst_texnum);
-       check_error();
-       glBindTexture(GL_TEXTURE_1D, dst_texnum);
+       glGenTextures(1, &src_texnum);
        check_error();
-       glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
+       glBindTexture(GL_TEXTURE_2D, src_texnum);
        check_error();
-       glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
+       glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
        check_error();
-       glTexImage1D(GL_TEXTURE_1D, 0, GL_LUMINANCE16F_ARB, 2, 0, GL_LUMINANCE, GL_FLOAT, texdata);
+       glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
        check_error();
-       glEnable(GL_TEXTURE_1D);
+       glTexImage2D(GL_TEXTURE_2D, 0, GL_R16F, 2, 1, 0, GL_RED, GL_FLOAT, texdata);
        check_error();
 
        // Basic state.
@@ -59,55 +77,195 @@ void measure_texel_subpixel_precision()
 
        glViewport(0, 0, width, 1);
 
-       glMatrixMode(GL_PROJECTION);
-       glLoadIdentity();
-       glOrtho(0.0, 1.0, 0.0, 1.0, 0.0, 1.0);
-
-       glMatrixMode(GL_MODELVIEW);
-       glLoadIdentity();
+       vector<string> frag_shader_outputs;
+       GLuint glsl_program_num = resource_pool.compile_glsl_program(
+               read_version_dependent_file("vs", "vert"),
+               read_version_dependent_file("texture1d", "frag"),
+               frag_shader_outputs);
+       glUseProgram(glsl_program_num);
+       check_error();
+       glUniform1i(glGetUniformLocation(glsl_program_num, "tex"), 0);  // Bind the 2D sampler.
        check_error();
 
        // Draw the texture stretched over a long quad, interpolating it out.
        // Note that since the texel center is in (0.5), we need to adjust the
        // texture coordinates in order not to get long stretches of (1,1,1,...)
        // at the start and (...,0,0,0) at the end.
-       glBegin(GL_QUADS);
+       float vertices[] = {
+               0.0f, 1.0f,
+               0.0f, 0.0f,
+               1.0f, 1.0f,
+               1.0f, 0.0f
+       };
+       float texcoords[] = {
+               0.25f, 0.0f,
+               0.25f, 0.0f,
+               0.75f, 0.0f,
+               0.75f, 0.0f
+       };
 
-       glTexCoord1f(0.25f);
-       glVertex2f(0.0f, 0.0f);
-
-       glTexCoord1f(0.75f);
-       glVertex2f(1.0f, 0.0f);
-
-       glTexCoord1f(0.75f);
-       glVertex2f(1.0f, 1.0f);
+       GLuint vao;
+       glGenVertexArrays(1, &vao);
+       check_error();
+       glBindVertexArray(vao);
+       check_error();
 
-       glTexCoord1f(0.25f);
-       glVertex2f(0.0f, 1.0f);
+       GLuint position_vbo = fill_vertex_attribute(glsl_program_num, "position", 2, GL_FLOAT, sizeof(vertices), vertices);
+       GLuint texcoord_vbo = fill_vertex_attribute(glsl_program_num, "texcoord", 2, GL_FLOAT, sizeof(texcoords), texcoords);
 
-       glEnd();
+       glDrawArrays(GL_TRIANGLE_STRIP, 0, 4);
        check_error();
 
-       glDisable(GL_TEXTURE_1D);
+       cleanup_vertex_attribute(glsl_program_num, "position", position_vbo);
+       cleanup_vertex_attribute(glsl_program_num, "texcoord", texcoord_vbo);
+
+       glUseProgram(0);
        check_error();
 
        // Now read the data back and see what the card did.
        // (We only look at the red channel; the others will surely be the same.)
        // We assume a linear ramp; anything else will give sort of odd results here.
-       float out_data[width];
-       glReadPixels(0, 0, width, 1, GL_RED, GL_FLOAT, out_data);
+       float out_data[width * 4];
+       glReadPixels(0, 0, width, 1, GL_RGBA, GL_FLOAT, out_data);
        check_error();
 
        float biggest_jump = 0.0f;
        for (unsigned i = 1; i < width; ++i) {
-               assert(out_data[i] >= out_data[i - 1]);
-               biggest_jump = std::max(biggest_jump, out_data[i] - out_data[i - 1]);
+               assert(out_data[i * 4] >= out_data[(i - 1) * 4]);
+               biggest_jump = max(biggest_jump, out_data[i * 4] - out_data[(i - 1) * 4]);
        }
 
+       assert(biggest_jump > 0.0);
        movit_texel_subpixel_precision = biggest_jump;
 
        // Clean up.
-       glBindTexture(GL_TEXTURE_1D, 0);
+       glBindTexture(GL_TEXTURE_2D, 0);
+       check_error();
+       glBindFramebuffer(GL_FRAMEBUFFER, 0);
+       check_error();
+       glDeleteFramebuffers(1, &fbo);
+       check_error();
+       glDeleteTextures(1, &dst_texnum);
+       check_error();
+       glDeleteTextures(1, &src_texnum);
+       check_error();
+
+       resource_pool.release_glsl_program(glsl_program_num);
+       glDeleteVertexArrays(1, &vao);
+       check_error();
+}
+
+void measure_roundoff_problems()
+{
+       ResourcePool resource_pool;
+
+       // Generate a destination texture to render to, and an FBO.
+       GLuint dst_texnum, fbo;
+
+       glGenTextures(1, &dst_texnum);
+       check_error();
+       glBindTexture(GL_TEXTURE_2D, dst_texnum);
+       check_error();
+       glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA8, 512, 1, 0, GL_RGBA, GL_UNSIGNED_BYTE, nullptr);
+       check_error();
+
+       glGenFramebuffers(1, &fbo);
+       check_error();
+       glBindFramebuffer(GL_FRAMEBUFFER, fbo);
+       check_error();
+       glFramebufferTexture2D(
+               GL_FRAMEBUFFER,
+               GL_COLOR_ATTACHMENT0,
+               GL_TEXTURE_2D,
+               dst_texnum,
+               0);
+       check_error();
+
+       // Now generate a texture where every value except the last should be
+       // rounded up to the next one. However, there are cards (in highly
+       // common use) that can't do this right, for unknown reasons.
+       GLuint src_texnum;
+       float texdata[512];
+       for (int i = 0; i < 256; ++i) {
+               texdata[i * 2 + 0] = (i + 0.48) / 255.0;
+               texdata[i * 2 + 1] = (i + 0.52) / 255.0;
+       }
+       glGenTextures(1, &src_texnum);
+       check_error();
+       glBindTexture(GL_TEXTURE_2D, src_texnum);
+       check_error();
+       glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
+       check_error();
+       glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
+       check_error();
+       glTexImage2D(GL_TEXTURE_2D, 0, GL_R32F, 512, 1, 0, GL_RED, GL_FLOAT, texdata);
+       check_error();
+
+       // Basic state.
+       glDisable(GL_BLEND);
+       check_error();
+       glDisable(GL_DEPTH_TEST);
+       check_error();
+       glDepthMask(GL_FALSE);
+       check_error();
+
+       glViewport(0, 0, 512, 1);
+
+       vector<string> frag_shader_outputs;
+       GLuint glsl_program_num = resource_pool.compile_glsl_program(
+               read_version_dependent_file("vs", "vert"),
+               read_version_dependent_file("texture1d", "frag"),
+               frag_shader_outputs);
+       glUseProgram(glsl_program_num);
+       check_error();
+       glUniform1i(glGetUniformLocation(glsl_program_num, "tex"), 0);  // Bind the 2D sampler.
+
+       // Draw the texture stretched over a long quad, interpolating it out.
+       float vertices[] = {
+               0.0f, 1.0f,
+               0.0f, 0.0f,
+               1.0f, 1.0f,
+               1.0f, 0.0f
+       };
+
+       GLuint vao;
+       glGenVertexArrays(1, &vao);
+       check_error();
+       glBindVertexArray(vao);
+       check_error();
+
+       GLuint position_vbo = fill_vertex_attribute(glsl_program_num, "position", 2, GL_FLOAT, sizeof(vertices), vertices);
+       GLuint texcoord_vbo = fill_vertex_attribute(glsl_program_num, "texcoord", 2, GL_FLOAT, sizeof(vertices), vertices);  // Same data.
+
+       glDrawArrays(GL_TRIANGLE_STRIP, 0, 4);
+       check_error();
+
+       cleanup_vertex_attribute(glsl_program_num, "position", position_vbo);
+       cleanup_vertex_attribute(glsl_program_num, "texcoord", texcoord_vbo);
+
+       glUseProgram(0);
+       check_error();
+
+       // Now read the data back and see what the card did. (Ignore the last value.)
+       // (We only look at the red channel; the others will surely be the same.)
+       unsigned char out_data[512 * 4];
+       glReadPixels(0, 0, 512, 1, GL_RGBA, GL_UNSIGNED_BYTE, out_data);
+       check_error();
+
+       int wrongly_rounded = 0;
+       for (unsigned i = 0; i < 255; ++i) {
+               if (out_data[(i * 2 + 0) * 4] != i) {
+                       ++wrongly_rounded;
+               }
+               if (out_data[(i * 2 + 1) * 4] != i + 1) {
+                       ++wrongly_rounded;
+               }
+       }
+
+       movit_num_wrongly_rounded = wrongly_rounded;
+
+       // Clean up.
+       glBindTexture(GL_TEXTURE_2D, 0);
        check_error();
        glBindFramebuffer(GL_FRAMEBUFFER, 0);
        check_error();
@@ -117,21 +275,159 @@ void measure_texel_subpixel_precision()
        check_error();
        glDeleteTextures(1, &src_texnum);
        check_error();
+
+       resource_pool.release_glsl_program(glsl_program_num);
+       glDeleteVertexArrays(1, &vao);
+       check_error();
+}
+
+bool check_extensions()
+{
+       // GLES generally doesn't use extensions as actively as desktop OpenGL.
+       // For now, we say that for GLES, we require GLES 3, which has everything
+       // we need.
+       if (!epoxy_is_desktop_gl()) {
+               if (epoxy_gl_version() >= 30) {
+                       return true;
+               } else {
+                       fprintf(stderr, "Movit system requirements: GLES version %.1f is too old (GLES 3.0 needed).\n",
+                               0.1f * epoxy_gl_version());
+                       fprintf(stderr, "Movit initialization failed.\n");
+                       return false;
+               }
+       }
+
+       if (epoxy_gl_version() < 30) {
+               fprintf(stderr, "Movit system requirements: OpenGL version %.1f is too old (OpenGL 3.0 needed).\n",
+                       0.1f * epoxy_gl_version());
+               fprintf(stderr, "Movit initialization failed.\n");
+               return false;
+       }
+
+       // The user can specify that they want a timing report for each
+       // phase in an effect chain. However, that depends on this extension;
+       // without it, we do cannot even create the query objects.
+       movit_timer_queries_supported =
+               (epoxy_gl_version() >= 33 || epoxy_has_gl_extension("GL_ARB_timer_query"));
+
+       // Certain effects have compute shader implementations, which may be
+       // more efficient than the normal fragment shader versions.
+       // GLSL 3.10 supposedly also has compute shaders, but I haven't tested them,
+       // so we require desktop OpenGL.
+       movit_compute_shaders_supported =
+               (epoxy_is_desktop_gl() &&
+                (epoxy_gl_version() >= 43 ||
+                 (epoxy_has_gl_extension("GL_ARB_compute_shader") &&
+                  epoxy_has_gl_extension("GL_ARB_shader_image_load_store") &&
+                  epoxy_has_gl_extension("GL_ARB_shader_image_size"))));
+
+       return true;
+}
+
+double get_glsl_version()
+{
+       char *glsl_version_str = strdup((const char *)glGetString(GL_SHADING_LANGUAGE_VERSION));
+
+       // Skip past the first period.
+       char *ptr = strchr(glsl_version_str, '.');
+       assert(ptr != nullptr);
+       ++ptr;
+
+       // Now cut the string off at the next period or space, whatever comes first
+       // (unless the string ends first).
+       while (*ptr && *ptr != '.' && *ptr != ' ') {
+               ++ptr;
+       }
+       *ptr = '\0';
+
+       // Now we have something on the form X.YY. We convert it to a float, and hope
+       // that if it's inexact (e.g. 1.30), atof() will round the same way the
+       // compiler will.
+       std::istringstream locale_convert(glsl_version_str);
+       locale_convert.imbue(std::locale("C"));
+       double glsl_version;
+       locale_convert >> glsl_version;
+       free(glsl_version_str);
+
+       return glsl_version;
+}
+
+void APIENTRY debug_callback(GLenum source,
+                             GLenum type,
+                             GLuint id,
+                             GLenum severity,
+                             GLsizei length,
+                             const char *message,
+                             const void *userParam)
+#ifdef __GNUC__
+       __attribute__((unused))
+#endif
+;
+
+void APIENTRY debug_callback(GLenum source,
+                             GLenum type,
+                             GLuint id,
+                             GLenum severity,
+                             GLsizei length,
+                             const char *message,
+                             const void *userParam)
+{
+       printf("Debug: %s\n", message);
 }
 
 }  // namespace
 
-void init_movit()
+bool init_movit(const string& data_directory, MovitDebugLevel debug_level)
 {
        if (movit_initialized) {
-               return;
+               return true;
        }
 
+       movit_data_directory = new string(data_directory);
+       movit_debug_level = debug_level;
+
        // geez 
        glPixelStorei(GL_PACK_ALIGNMENT, 1);
        glPixelStorei(GL_UNPACK_ALIGNMENT, 1);
+       glDisable(GL_DITHER);
+
+       // You can turn this on if you want detailed debug messages from the driver.
+       // You should probably also ask for a debug context (see gtest_sdl_main.cpp),
+       // or you might not get much data back.
+       // glDebugMessageCallbackARB(callback, nullptr);
+       // glDebugMessageControlARB(GL_DONT_CARE, GL_DONT_CARE, GL_DONT_CARE, 0, 0, GL_TRUE);
+
+       if (!check_extensions()) {
+               return false;
+       }
+
+       // Find out what shader model we should compile for.
+       // We need at least 1.30, due to use of (among others) integers.
+       if (epoxy_is_desktop_gl()) {
+               if (get_glsl_version() < 1.30f) {
+                       fprintf(stderr, "Movit system requirements: Needs at least GLSL version 1.30 (has version %.1f)\n",
+                               get_glsl_version());
+                       return false;
+               }
+               if (get_glsl_version() < 1.50f) {
+                       movit_shader_model = MOVIT_GLSL_130;
+               } else {
+                       // Note: All of our 1.50 shaders are identical to our 1.30 shaders,
+                       // but OS X does not support 1.30; only 1.10 (which we don't support
+                       // anymore) and 1.50 (and then only with core contexts). So we keep
+                       // a second set of shaders around whose only difference is the different
+                       // #version declaration.
+                       movit_shader_model = MOVIT_GLSL_150;
+               }
+       } else {
+               movit_shader_model = MOVIT_ESSL_300;
+       }
 
        measure_texel_subpixel_precision();
+       measure_roundoff_problems();
 
        movit_initialized = true;
+       return true;
 }
+
+}  // namespace movit