X-Git-Url: https://git.sesse.net/?p=movit;a=blobdiff_plain;f=init.cpp;h=5e6c6427e45d7155ab2032652e0af554010be115;hp=4003163d9494843b3eccc42910ca41084b52a691;hb=eff011224abc5dc81f801f3ea44572287a55bcac;hpb=92573f7be18bfca16dd3292cbe536d7d7912198f diff --git a/init.cpp b/init.cpp index 4003163..5e6c642 100644 --- a/init.cpp +++ b/init.cpp @@ -1,4 +1,4 @@ -#include +#include #include #include #include @@ -15,14 +15,14 @@ namespace movit { bool movit_initialized = false; MovitDebugLevel movit_debug_level = MOVIT_DEBUG_ON; float movit_texel_subpixel_precision; -bool movit_srgb_textures_supported; +bool movit_timer_queries_supported, movit_compute_shaders_supported; int movit_num_wrongly_rounded; -bool movit_shader_rounding_supported; +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 = NULL; +string *movit_data_directory = nullptr; namespace { @@ -38,7 +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); @@ -58,13 +58,13 @@ void measure_texel_subpixel_precision() float texdata[] = { 0, 1 }; glGenTextures(1, &src_texnum); check_error(); - glBindTexture(GL_TEXTURE_1D, src_texnum); + glBindTexture(GL_TEXTURE_2D, src_texnum); check_error(); - glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MIN_FILTER, GL_LINEAR); + glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR); check_error(); - glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE); + glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE); check_error(); - glTexImage1D(GL_TEXTURE_1D, 0, GL_LUMINANCE16F_ARB, 2, 0, GL_LUMINANCE, GL_FLOAT, texdata); + glTexImage2D(GL_TEXTURE_2D, 0, GL_R16F, 2, 1, 0, GL_RED, GL_FLOAT, texdata); check_error(); // Basic state. @@ -77,11 +77,14 @@ void measure_texel_subpixel_precision() glViewport(0, 0, width, 1); + vector frag_shader_outputs; GLuint glsl_program_num = resource_pool.compile_glsl_program( - read_file("vs.vert"), read_file("texture1d.frag")); + 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 1D sampler. + glUniform1i(glGetUniformLocation(glsl_program_num, "tex"), 0); // Bind the 2D sampler. check_error(); // Draw the texture stretched over a long quad, interpolating it out. @@ -89,59 +92,54 @@ void measure_texel_subpixel_precision() // texture coordinates in order not to get long stretches of (1,1,1,...) // at the start and (...,0,0,0) at the end. float vertices[] = { + 0.0f, 1.0f, 0.0f, 0.0f, - 1.0f, 0.0f, 1.0f, 1.0f, - 0.0f, 1.0f + 1.0f, 0.0f }; float texcoords[] = { + 0.25f, 0.0f, 0.25f, 0.0f, 0.75f, 0.0f, - 0.75f, 0.0f, - 0.25f, 0.0f + 0.75f, 0.0f }; - int position_attrib = glGetAttribLocation(glsl_program_num, "position"); - assert(position_attrib != -1); - int texcoord_attrib = glGetAttribLocation(glsl_program_num, "texcoord"); - assert(texcoord_attrib != -1); - glEnableVertexAttribArray(position_attrib); - check_error(); - glVertexAttribPointer(position_attrib, 2, GL_FLOAT, GL_FALSE, 0, vertices); + GLuint vao; + glGenVertexArrays(1, &vao); check_error(); - glEnableVertexAttribArray(texcoord_attrib); - check_error(); - glVertexAttribPointer(texcoord_attrib, 2, GL_FLOAT, GL_FALSE, 0, texcoords); + glBindVertexArray(vao); check_error(); - glDrawArrays(GL_QUADS, 0, 4); + 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); + + 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(); - glDisableVertexAttribArray(position_attrib); - check_error(); - glDisableVertexAttribArray(texcoord_attrib); - 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 = 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(); @@ -153,6 +151,8 @@ void measure_texel_subpixel_precision() check_error(); resource_pool.release_glsl_program(glsl_program_num); + glDeleteVertexArrays(1, &vao); + check_error(); } void measure_roundoff_problems() @@ -166,7 +166,7 @@ void measure_roundoff_problems() 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, NULL); + glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA8, 512, 1, 0, GL_RGBA, GL_UNSIGNED_BYTE, nullptr); check_error(); glGenFramebuffers(1, &fbo); @@ -192,13 +192,13 @@ void measure_roundoff_problems() } glGenTextures(1, &src_texnum); check_error(); - glBindTexture(GL_TEXTURE_1D, src_texnum); + glBindTexture(GL_TEXTURE_2D, src_texnum); check_error(); - glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MIN_FILTER, GL_LINEAR); + glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR); check_error(); - glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE); + glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE); check_error(); - glTexImage1D(GL_TEXTURE_1D, 0, GL_LUMINANCE32F_ARB, 512, 0, GL_LUMINANCE, GL_FLOAT, texdata); + glTexImage2D(GL_TEXTURE_2D, 0, GL_R32F, 512, 1, 0, GL_RED, GL_FLOAT, texdata); check_error(); // Basic state. @@ -211,57 +211,53 @@ void measure_roundoff_problems() glViewport(0, 0, 512, 1); + vector frag_shader_outputs; GLuint glsl_program_num = resource_pool.compile_glsl_program( - read_file("vs.vert"), read_file("texture1d.frag")); + 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 1D sampler. + 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, 0.0f, 1.0f, 1.0f, - 0.0f, 1.0f + 1.0f, 0.0f }; - float texcoords[] = { - 0.25f, 0.0f, - 0.75f, 0.0f, - 0.75f, 0.0f, - 0.25f, 0.0f - }; - int position_attrib = glGetAttribLocation(glsl_program_num, "position"); - assert(position_attrib != -1); - int texcoord_attrib = glGetAttribLocation(glsl_program_num, "texcoord"); - assert(texcoord_attrib != -1); - glEnableVertexAttribArray(position_attrib); - check_error(); - glVertexAttribPointer(position_attrib, 2, GL_FLOAT, GL_FALSE, 0, vertices); + + GLuint vao; + glGenVertexArrays(1, &vao); check_error(); - glEnableVertexAttribArray(texcoord_attrib); + glBindVertexArray(vao); check_error(); - glVertexAttribPointer(texcoord_attrib, 2, GL_FLOAT, GL_FALSE, 0, texcoords); + + 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(); - glDrawArrays(GL_QUADS, 0, 4); + + cleanup_vertex_attribute(glsl_program_num, "position", position_vbo); + cleanup_vertex_attribute(glsl_program_num, "texcoord", texcoord_vbo); glUseProgram(0); check_error(); - glDisableVertexAttribArray(position_attrib); - check_error(); - glDisableVertexAttribArray(texcoord_attrib); // 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]; - glReadPixels(0, 0, 512, 1, GL_RED, GL_UNSIGNED_BYTE, out_data); + 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] != i) { + if (out_data[(i * 2 + 0) * 4] != i) { ++wrongly_rounded; } - if (out_data[i * 2 + 1] != i + 1) { + if (out_data[(i * 2 + 1) * 4] != i + 1) { ++wrongly_rounded; } } @@ -269,7 +265,7 @@ void measure_roundoff_problems() movit_num_wrongly_rounded = wrongly_rounded; // Clean up. - glBindTexture(GL_TEXTURE_1D, 0); + glBindTexture(GL_TEXTURE_2D, 0); check_error(); glBindFramebuffer(GL_FRAMEBUFFER, 0); check_error(); @@ -281,39 +277,102 @@ void measure_roundoff_problems() check_error(); resource_pool.release_glsl_program(glsl_program_num); + glDeleteVertexArrays(1, &vao); + check_error(); } bool check_extensions() { - // We fundamentally need FBOs and floating-point textures. - if (!glewIsSupported("GL_ARB_framebuffer_object")) return false; - if (!glewIsSupported("GL_ARB_texture_float")) return false; + // 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; + } + } - // We assume that we can use non-power-of-two textures without restrictions. - if (!glewIsSupported("GL_ARB_texture_non_power_of_two")) 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; + } - // We also need GLSL fragment shaders. - if (!glewIsSupported("GL_ARB_fragment_shader")) return false; - if (!glewIsSupported("GL_ARB_shading_language_100")) 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")))); - // FlatInput and YCbCrInput uses PBOs. (They could in theory do without, - // but no modern card would really not provide it.) - if (!glewIsSupported("GL_ARB_pixel_buffer_object")) return false; + return true; +} - // ResampleEffect uses RG textures to encode a two-component LUT. - if (!glewIsSupported("GL_ARB_texture_rg")) return false; +double get_glsl_version() +{ + char *glsl_version_str = strdup((const char *)glGetString(GL_SHADING_LANGUAGE_VERSION)); - // sRGB texture decode would be nice, but are not mandatory - // (GammaExpansionEffect can do the same thing if needed). - movit_srgb_textures_supported = glewIsSupported("GL_EXT_texture_sRGB"); + // Skip past the first period. + char *ptr = strchr(glsl_version_str, '.'); + assert(ptr != nullptr); + ++ptr; - // We may want to use round() at the end of the final shader, - // if supported. We need either GLSL 1.30 or this extension to do that, - // and 1.30 brings with it other things that we don't want to demand - // for now. - movit_shader_rounding_supported = glewIsSupported("GL_EXT_gpu_shader4"); + // 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; +} - return true; +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 @@ -327,19 +386,43 @@ bool init_movit(const string& data_directory, MovitDebugLevel debug_level) movit_data_directory = new string(data_directory); movit_debug_level = debug_level; - GLenum err = glewInit(); - if (err != GLEW_OK) { - return false; - } - // 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();