]> git.sesse.net Git - movit/blobdiff - effect_chain.cpp
Bump version after new format support.
[movit] / effect_chain.cpp
index c2402daeaef400d9ded80dcbce3378be9f94ac42..fa7340d7a456839f7ef2563d76c5a872182491ff 100644 (file)
@@ -1,5 +1,3 @@
-#define GL_GLEXT_PROTOTYPES 1
-
 #include <epoxy/gl.h>
 #include <assert.h>
 #include <math.h>
@@ -34,12 +32,13 @@ using namespace std;
 
 namespace movit {
 
-EffectChain::EffectChain(float aspect_nom, float aspect_denom, ResourcePool *resource_pool)
+EffectChain::EffectChain(float aspect_nom, float aspect_denom, ResourcePool *resource_pool, GLenum intermediate_format)
        : aspect_nom(aspect_nom),
          aspect_denom(aspect_denom),
          output_color_rgba(false),
          output_color_ycbcr(false),
          dither_effect(NULL),
+         intermediate_format(intermediate_format),
          num_dither_bits(0),
          output_origin(OUTPUT_ORIGIN_BOTTOM_LEFT),
          finalized(false),
@@ -51,6 +50,14 @@ EffectChain::EffectChain(float aspect_nom, float aspect_denom, ResourcePool *res
        } else {
                owns_resource_pool = false;
        }
+
+       // Generate a VBO with some data in (shared position and texture coordinate data).
+       float vertices[] = {
+               0.0f, 2.0f,
+               0.0f, 0.0f,
+               2.0f, 0.0f
+       };
+       vbo = generate_vbo(2, GL_FLOAT, sizeof(vertices), vertices);
 }
 
 EffectChain::~EffectChain()
@@ -66,6 +73,8 @@ EffectChain::~EffectChain()
        if (owns_resource_pool) {
                delete resource_pool;
        }
+       glDeleteBuffers(1, &vbo);
+       check_error();
 }
 
 Input *EffectChain::add_input(Input *input)
@@ -448,6 +457,14 @@ void EffectChain::compile_glsl_program(Phase *phase)
        }
 
        phase->glsl_program_num = resource_pool->compile_glsl_program(vert_shader, frag_shader, frag_shader_outputs);
+       GLint position_attribute_index = glGetAttribLocation(phase->glsl_program_num, "position");
+       GLint texcoord_attribute_index = glGetAttribLocation(phase->glsl_program_num, "texcoord");
+       if (position_attribute_index != -1) {
+               phase->attribute_indexes.insert(position_attribute_index);
+       }
+       if (texcoord_attribute_index != -1) {
+               phase->attribute_indexes.insert(texcoord_attribute_index);
+       }
 
        // Collect the resulting location numbers for each uniform.
        collect_uniform_locations(phase->glsl_program_num, &phase->uniforms_sampler2d);
@@ -589,9 +606,17 @@ Phase *EffectChain::construct_phase(Node *output, map<Node *, Phase *> *complete
        // and create a GLSL program for it.
        assert(!phase->effects.empty());
 
-       // Deduplicate the inputs.
-       sort(phase->inputs.begin(), phase->inputs.end());
-       phase->inputs.erase(unique(phase->inputs.begin(), phase->inputs.end()), phase->inputs.end());
+       // Deduplicate the inputs, but don't change the ordering e.g. by sorting;
+       // that would be nondeterministic and thus reduce cacheability.
+       // TODO: Make this even more deterministic.
+       vector<Phase *> dedup_inputs;
+       set<Phase *> seen_inputs;
+       for (size_t i = 0; i < phase->inputs.size(); ++i) {
+               if (seen_inputs.insert(phase->inputs[i]).second) {
+                       dedup_inputs.push_back(phase->inputs[i]);
+               }
+       }
+       swap(phase->inputs, dedup_inputs);
 
        // Allocate samplers for each input.
        phase->input_samplers.resize(phase->inputs.size());
@@ -626,9 +651,8 @@ Phase *EffectChain::construct_phase(Node *output, map<Node *, Phase *> *complete
        // Actually make the shader for this phase.
        compile_glsl_program(phase);
 
-       // Initialize timer objects.
+       // Initialize timers.
        if (movit_timer_queries_supported) {
-               glGenQueries(1, &phase->timer_query_object);
                phase->time_elapsed_ns = 0;
                phase->num_measured_iterations = 0;
        }
@@ -1669,6 +1693,8 @@ void EffectChain::render_to_fbo(GLuint dest_fbo, unsigned width, unsigned height
        check_error();
        glDisable(GL_DITHER);
        check_error();
+       glEnable(GL_FRAMEBUFFER_SRGB);
+       check_error();
 
        // Save original viewport.
        GLuint x = 0, y = 0;
@@ -1691,22 +1717,16 @@ void EffectChain::render_to_fbo(GLuint dest_fbo, unsigned width, unsigned height
        glDepthMask(GL_FALSE);
        check_error();
 
-       // Generate a VAO. All the phases should have exactly the same vertex attributes,
-       // so it's safe to reuse this.
-       float vertices[] = {
-               0.0f, 2.0f,
-               0.0f, 0.0f,
-               2.0f, 0.0f
-       };
-
+       // Generate a VAO that will be used during the entire execution,
+       // and bind the VBO, since it contains all the data.
        GLuint vao;
        glGenVertexArrays(1, &vao);
        check_error();
        glBindVertexArray(vao);
        check_error();
-
-       GLuint position_vbo = fill_vertex_attribute(phases[0]->glsl_program_num, "position", 2, GL_FLOAT, sizeof(vertices), vertices);
-       GLuint texcoord_vbo = fill_vertex_attribute(phases[0]->glsl_program_num, "texcoord", 2, GL_FLOAT, sizeof(vertices), vertices);  // Same as vertices.
+       glBindBuffer(GL_ARRAY_BUFFER, vbo);
+       check_error();
+       set<GLint> bound_attribute_indices;
 
        set<Phase *> generated_mipmaps;
 
@@ -1718,7 +1738,15 @@ void EffectChain::render_to_fbo(GLuint dest_fbo, unsigned width, unsigned height
                Phase *phase = phases[phase_num];
 
                if (do_phase_timing) {
-                       glBeginQuery(GL_TIME_ELAPSED, phase->timer_query_object);
+                       GLuint timer_query_object;
+                       if (phase->timer_query_objects_free.empty()) {
+                               glGenQueries(1, &timer_query_object);
+                       } else {
+                               timer_query_object = phase->timer_query_objects_free.front();
+                               phase->timer_query_objects_free.pop_front();
+                       }
+                       glBeginQuery(GL_TIME_ELAPSED, timer_query_object);
+                       phase->timer_query_objects_running.push_back(timer_query_object);
                }
                if (phase_num == phases.size() - 1) {
                        // Last phase goes to the output the user specified.
@@ -1732,7 +1760,7 @@ void EffectChain::render_to_fbo(GLuint dest_fbo, unsigned width, unsigned height
                                CHECK(dither_effect->set_int("output_height", height));
                        }
                }
-               execute_phase(phase, phase_num == phases.size() - 1, &output_textures, &generated_mipmaps);
+               execute_phase(phase, phase_num == phases.size() - 1, &bound_attribute_indices, &output_textures, &generated_mipmaps);
                if (do_phase_timing) {
                        glEndQuery(GL_TIME_ELAPSED);
                }
@@ -1749,9 +1777,10 @@ void EffectChain::render_to_fbo(GLuint dest_fbo, unsigned width, unsigned height
        glUseProgram(0);
        check_error();
 
-       cleanup_vertex_attribute(phases[0]->glsl_program_num, "position", position_vbo);
-       cleanup_vertex_attribute(phases[0]->glsl_program_num, "texcoord", texcoord_vbo);
-
+       glBindBuffer(GL_ARRAY_BUFFER, 0);
+       check_error();
+       glBindVertexArray(0);
+       check_error();
        glDeleteVertexArrays(1, &vao);
        check_error();
 
@@ -1759,14 +1788,22 @@ void EffectChain::render_to_fbo(GLuint dest_fbo, unsigned width, unsigned height
                // Get back the timer queries.
                for (unsigned phase_num = 0; phase_num < phases.size(); ++phase_num) {
                        Phase *phase = phases[phase_num];
-                       GLint available = 0;
-                       while (!available) {
-                               glGetQueryObjectiv(phase->timer_query_object, GL_QUERY_RESULT_AVAILABLE, &available);
+                       for (std::list<GLuint>::iterator timer_it = phase->timer_query_objects_running.begin();
+                            timer_it != phase->timer_query_objects_running.end(); ) {
+                               GLint timer_query_object = *timer_it;
+                               GLint available;
+                               glGetQueryObjectiv(timer_query_object, GL_QUERY_RESULT_AVAILABLE, &available);
+                               if (available) {
+                                       GLuint64 time_elapsed;
+                                       glGetQueryObjectui64v(timer_query_object, GL_QUERY_RESULT, &time_elapsed);
+                                       phase->time_elapsed_ns += time_elapsed;
+                                       ++phase->num_measured_iterations;
+                                       phase->timer_query_objects_free.push_back(timer_query_object);
+                                       phase->timer_query_objects_running.erase(timer_it++);
+                               } else {
+                                       ++timer_it;
+                               }
                        }
-                       GLuint64 time_elapsed;
-                       glGetQueryObjectui64v(phase->timer_query_object, GL_QUERY_RESULT, &time_elapsed);
-                       phase->time_elapsed_ns += time_elapsed;
-                       ++phase->num_measured_iterations;
                }
        }
 }
@@ -1807,7 +1844,10 @@ void EffectChain::print_phase_timing()
        printf("Total:   %5.1f ms\n", total_time_ms);
 }
 
-void EffectChain::execute_phase(Phase *phase, bool last_phase, map<Phase *, GLuint> *output_textures, set<Phase *> *generated_mipmaps)
+void EffectChain::execute_phase(Phase *phase, bool last_phase,
+                                set<GLint> *bound_attribute_indices,
+                                map<Phase *, GLuint> *output_textures,
+                                set<Phase *> *generated_mipmaps)
 {
        GLuint fbo = 0;
 
@@ -1816,13 +1856,11 @@ void EffectChain::execute_phase(Phase *phase, bool last_phase, map<Phase *, GLui
        if (!last_phase) {
                find_output_size(phase);
 
-               GLuint tex_num = resource_pool->create_2d_texture(GL_RGBA16F, phase->output_width, phase->output_height);
+               GLuint tex_num = resource_pool->create_2d_texture(intermediate_format, phase->output_width, phase->output_height);
                output_textures->insert(make_pair(phase, tex_num));
        }
 
-       const GLuint glsl_program_num = phase->glsl_program_num;
-       check_error();
-       glUseProgram(glsl_program_num);
+       glUseProgram(phase->glsl_program_num);
        check_error();
 
        // Set up RTT inputs for this phase.
@@ -1853,7 +1891,7 @@ void EffectChain::execute_phase(Phase *phase, bool last_phase, map<Phase *, GLui
        for (unsigned i = 0; i < phase->effects.size(); ++i) {
                Node *node = phase->effects[i];
                unsigned old_sampler_num = sampler_num;
-               node->effect->set_gl_state(glsl_program_num, phase->effect_ids[node], &sampler_num);
+               node->effect->set_gl_state(phase->glsl_program_num, phase->effect_ids[node], &sampler_num);
                check_error();
 
                if (node->effect->is_single_texture()) {
@@ -1868,12 +1906,34 @@ void EffectChain::execute_phase(Phase *phase, bool last_phase, map<Phase *, GLui
        // from there.
        setup_uniforms(phase);
 
-       glDrawArrays(GL_TRIANGLES, 0, 3);
-       check_error();
+       // Clean up old attributes if they are no longer needed.
+       for (set<GLint>::iterator attr_it = bound_attribute_indices->begin();
+            attr_it != bound_attribute_indices->end(); ) {
+               if (phase->attribute_indexes.count(*attr_it) == 0) {
+                       glDisableVertexAttribArray(*attr_it);
+                       check_error();
+                       bound_attribute_indices->erase(attr_it++);
+               } else {
+                       ++attr_it;
+               }
+       }
 
-       glUseProgram(0);
-       check_error();
+       // Set up the new attributes, if needed.
+       for (set<GLint>::iterator attr_it = phase->attribute_indexes.begin();
+            attr_it != phase->attribute_indexes.end();
+            ++attr_it) {
+               if (bound_attribute_indices->count(*attr_it) == 0) {
+                       glEnableVertexAttribArray(*attr_it);
+                       check_error();
+                       glVertexAttribPointer(*attr_it, 2, GL_FLOAT, GL_FALSE, 0, BUFFER_OFFSET(0));
+                       check_error();
+                       bound_attribute_indices->insert(*attr_it);
+               }
+       }
 
+       glDrawArrays(GL_TRIANGLES, 0, 3);
+       check_error();
+       
        for (unsigned i = 0; i < phase->effects.size(); ++i) {
                Node *node = phase->effects[i];
                node->effect->clear_gl_state();