]> git.sesse.net Git - movit/blobdiff - effect_chain.cpp
Correct/update a comment on rounding.
[movit] / effect_chain.cpp
index bf017166a770f5a2d4d52adf0d68cf303aab30bd..5e43474c457e8cc7dca3db7054fb8e6b6581e3d8 100644 (file)
@@ -1,5 +1,3 @@
-#define GL_GLEXT_PROTOTYPES 1
-
 #include <epoxy/gl.h>
 #include <assert.h>
 #include <math.h>
@@ -37,8 +35,11 @@ namespace movit {
 EffectChain::EffectChain(float aspect_nom, float aspect_denom, ResourcePool *resource_pool)
        : aspect_nom(aspect_nom),
          aspect_denom(aspect_denom),
+         output_color_rgba(false),
+         output_color_ycbcr(false),
          dither_effect(NULL),
          num_dither_bits(0),
+         output_origin(OUTPUT_ORIGIN_BOTTOM_LEFT),
          finalized(false),
          resource_pool(resource_pool),
          do_phase_timing(false) {
@@ -48,6 +49,14 @@ 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()
@@ -63,6 +72,8 @@ EffectChain::~EffectChain()
        if (owns_resource_pool) {
                delete resource_pool;
        }
+       glDeleteBuffers(1, &vbo);
+       check_error();
 }
 
 Input *EffectChain::add_input(Input *input)
@@ -76,19 +87,22 @@ Input *EffectChain::add_input(Input *input)
 void EffectChain::add_output(const ImageFormat &format, OutputAlphaFormat alpha_format)
 {
        assert(!finalized);
+       assert(!output_color_rgba);
        output_format = format;
        output_alpha_format = alpha_format;
-       output_color_type = OUTPUT_COLOR_RGB;
+       output_color_rgba = true;
 }
 
 void EffectChain::add_ycbcr_output(const ImageFormat &format, OutputAlphaFormat alpha_format,
-                                   const YCbCrFormat &ycbcr_format)
+                                   const YCbCrFormat &ycbcr_format, YCbCrOutputSplitting output_splitting)
 {
        assert(!finalized);
+       assert(!output_color_ycbcr);
        output_format = format;
        output_alpha_format = alpha_format;
-       output_color_type = OUTPUT_COLOR_YCBCR;
+       output_color_ycbcr = true;
        output_ycbcr_format = ycbcr_format;
+       output_ycbcr_splitting = output_splitting;
 
        assert(ycbcr_format.chroma_subsampling_x == 1);
        assert(ycbcr_format.chroma_subsampling_y == 1);
@@ -178,6 +192,13 @@ GLenum EffectChain::get_input_sampler(Node *node, unsigned input_num) const
        return GL_TEXTURE0 + node->incoming_links[input_num]->bound_sampler_num;
 }
 
+GLenum EffectChain::has_input_sampler(Node *node, unsigned input_num) const
+{
+       assert(input_num < node->incoming_links.size());
+       return node->incoming_links[input_num]->bound_sampler_num >= 0 &&
+               node->incoming_links[input_num]->bound_sampler_num < 8;
+}
+
 void EffectChain::find_all_nonlinear_inputs(Node *node, vector<Node *> *nonlinear_inputs)
 {
        if (node->output_gamma_curve == GAMMA_LINEAR &&
@@ -206,7 +227,7 @@ Effect *EffectChain::add_effect(Effect *effect, const vector<Effect *> &inputs)
        return effect;
 }
 
-// GLSL pre-1.30 doesn't support token pasting. Replace PREFIX(x) with <effect_id>_x.
+// ESSL doesn't support token pasting. Replace PREFIX(x) with <effect_id>_x.
 string replace_prefix(const string &text, const string &prefix)
 {
        string output;
@@ -361,7 +382,39 @@ void EffectChain::compile_glsl_program(Phase *phase)
                frag_shader += "\n";
        }
        frag_shader += string("#define INPUT ") + phase->effect_ids[phase->effects.back()] + "\n";
-       frag_shader.append(read_version_dependent_file("footer", "frag"));
+
+       // If we're the last phase, add the right #defines for Y'CbCr multi-output as needed.
+       vector<string> frag_shader_outputs;  // In order.
+       if (phase->output_node->outgoing_links.empty() && output_color_ycbcr) {
+               switch (output_ycbcr_splitting) {
+               case YCBCR_OUTPUT_INTERLEAVED:
+                       // No #defines set.
+                       frag_shader_outputs.push_back("FragColor");
+                       break;
+               case YCBCR_OUTPUT_SPLIT_Y_AND_CBCR:
+                       frag_shader += "#define YCBCR_OUTPUT_SPLIT_Y_AND_CBCR 1\n";
+                       frag_shader_outputs.push_back("Y");
+                       frag_shader_outputs.push_back("Chroma");
+                       break;
+               case YCBCR_OUTPUT_PLANAR:
+                       frag_shader += "#define YCBCR_OUTPUT_PLANAR 1\n";
+                       frag_shader_outputs.push_back("Y");
+                       frag_shader_outputs.push_back("Cb");
+                       frag_shader_outputs.push_back("Cr");
+                       break;
+               default:
+                       assert(false);
+               }
+
+               if (output_color_rgba) {
+                       // Note: Needs to come in the header, because not only the
+                       // output needs to see it (YCbCrConversionEffect and DitherEffect
+                       // do, too).
+                       frag_shader_header += "#define YCBCR_ALSO_OUTPUT_RGBA 1\n";
+                       frag_shader_outputs.push_back("RGBA");
+               }
+       }
+       frag_shader.append(read_file("footer.frag"));
 
        // Collect uniforms from all effects and output them. Note that this needs
        // to happen after output_fragment_shader(), even though the uniforms come
@@ -381,7 +434,9 @@ void EffectChain::compile_glsl_program(Phase *phase)
                extract_uniform_declarations(effect->uniforms_vec2, "vec2", effect_id, &phase->uniforms_vec2, &frag_shader_uniforms);
                extract_uniform_declarations(effect->uniforms_vec3, "vec3", effect_id, &phase->uniforms_vec3, &frag_shader_uniforms);
                extract_uniform_declarations(effect->uniforms_vec4, "vec4", effect_id, &phase->uniforms_vec4, &frag_shader_uniforms);
+               extract_uniform_array_declarations(effect->uniforms_float_array, "float", effect_id, &phase->uniforms_float, &frag_shader_uniforms);
                extract_uniform_array_declarations(effect->uniforms_vec2_array, "vec2", effect_id, &phase->uniforms_vec2, &frag_shader_uniforms);
+               extract_uniform_array_declarations(effect->uniforms_vec3_array, "vec3", effect_id, &phase->uniforms_vec3, &frag_shader_uniforms);
                extract_uniform_array_declarations(effect->uniforms_vec4_array, "vec4", effect_id, &phase->uniforms_vec4, &frag_shader_uniforms);
                extract_uniform_declarations(effect->uniforms_mat3, "mat3", effect_id, &phase->uniforms_mat3, &frag_shader_uniforms);
        }
@@ -389,7 +444,26 @@ void EffectChain::compile_glsl_program(Phase *phase)
        frag_shader = frag_shader_header + frag_shader_uniforms + frag_shader;
 
        string vert_shader = read_version_dependent_file("vs", "vert");
-       phase->glsl_program_num = resource_pool->compile_glsl_program(vert_shader, frag_shader);
+
+       // If we're the last phase and need to flip the picture to compensate for
+       // the origin, tell the vertex shader so.
+       if (phase->output_node->outgoing_links.empty() && output_origin == OUTPUT_ORIGIN_TOP_LEFT) {
+               const string needle = "#define FLIP_ORIGIN 0";
+               size_t pos = vert_shader.find(needle);
+               assert(pos != string::npos);
+
+               vert_shader[pos + needle.size() - 1] = '1';
+       }
+
+       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);
@@ -459,7 +533,8 @@ Phase *EffectChain::construct_phase(Node *output, map<Node *, Phase *> *complete
                        bool start_new_phase = false;
 
                        if (node->effect->needs_texture_bounce() &&
-                           !deps[i]->effect->is_single_texture()) {
+                           !deps[i]->effect->is_single_texture() &&
+                           !deps[i]->effect->override_disable_bounce()) {
                                start_new_phase = true;
                        }
 
@@ -530,9 +605,17 @@ 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());
@@ -1489,8 +1572,8 @@ void EffectChain::fix_output_gamma()
 // gamma-encoded data.
 void EffectChain::add_ycbcr_conversion_if_needed()
 {
-       assert(output_color_type == OUTPUT_COLOR_RGB || output_color_type == OUTPUT_COLOR_YCBCR);
-       if (output_color_type != OUTPUT_COLOR_YCBCR) {
+       assert(output_color_rgba || output_color_ycbcr);
+       if (!output_color_ycbcr) {
                return;
        }
        Node *output = find_output_node();
@@ -1605,6 +1688,12 @@ void EffectChain::render_to_fbo(GLuint dest_fbo, unsigned width, unsigned height
 {
        assert(finalized);
 
+       // This needs to be set anew, in case we are coming from a different context
+       // from when we initialized.
+       check_error();
+       glDisable(GL_DITHER);
+       check_error();
+
        // Save original viewport.
        GLuint x = 0, y = 0;
 
@@ -1618,6 +1707,7 @@ void EffectChain::render_to_fbo(GLuint dest_fbo, unsigned width, unsigned height
        }
 
        // Basic state.
+       check_error();
        glDisable(GL_BLEND);
        check_error();
        glDisable(GL_DEPTH_TEST);
@@ -1625,22 +1715,16 @@ 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;
 
@@ -1666,7 +1750,7 @@ 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);
                }
@@ -1683,9 +1767,10 @@ 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();
 
@@ -1741,7 +1826,10 @@ 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;
 
@@ -1754,9 +1842,7 @@ void EffectChain::execute_phase(Phase *phase, bool last_phase, map<Phase *, GLui
                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.
@@ -1787,7 +1873,7 @@ 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()) {
@@ -1802,12 +1888,34 @@ 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();