X-Git-Url: https://git.sesse.net/?p=movit;a=blobdiff_plain;f=effect_chain.cpp;h=9b77ef032f49f5359df6fb1925852b1b9b6bc610;hp=8a72e1756dd90a98502de0944ffc00225a676227;hb=c1bd648471212ae960a26c1fb518223416a83f4b;hpb=20c2ca726f0e6c194a1fddcb9908796de19430ac diff --git a/effect_chain.cpp b/effect_chain.cpp index 8a72e17..9b77ef0 100644 --- a/effect_chain.cpp +++ b/effect_chain.cpp @@ -51,6 +51,14 @@ 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 +74,8 @@ EffectChain::~EffectChain() if (owns_resource_pool) { delete resource_pool; } + glDeleteBuffers(1, &vbo); + check_error(); } Input *EffectChain::add_input(Input *input) @@ -376,16 +386,23 @@ void EffectChain::compile_glsl_program(Phase *phase) frag_shader += string("#define INPUT ") + phase->effect_ids[phase->effects.back()] + "\n"; // If we're the last phase, add the right #defines for Y'CbCr multi-output as needed. + vector 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); @@ -396,6 +413,7 @@ void EffectChain::compile_glsl_program(Phase *phase) // 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")); @@ -439,7 +457,15 @@ void EffectChain::compile_glsl_program(Phase *phase) vert_shader[pos + needle.size() - 1] = '1'; } - phase->glsl_program_num = resource_pool->compile_glsl_program(vert_shader, frag_shader); + 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); @@ -581,9 +607,17 @@ Phase *EffectChain::construct_phase(Node *output, map *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 dedup_inputs; + set 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()); @@ -1683,22 +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 bound_attribute_indices; set generated_mipmaps; @@ -1724,12 +1752,19 @@ 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); } } + for (set::iterator attr_it = bound_attribute_indices.begin(); + attr_it != bound_attribute_indices.end(); + ++attr_it) { + glDisableVertexAttribArray(*attr_it); + check_error(); + } + for (map::const_iterator texture_it = output_textures.begin(); texture_it != output_textures.end(); ++texture_it) { @@ -1741,9 +1776,10 @@ 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(); @@ -1799,7 +1835,10 @@ void EffectChain::print_phase_timing() printf("Total: %5.1f ms\n", total_time_ms); } -void EffectChain::execute_phase(Phase *phase, bool last_phase, map *output_textures, set *generated_mipmaps) +void EffectChain::execute_phase(Phase *phase, bool last_phase, + set *bound_attribute_indices, + map *output_textures, + set *generated_mipmaps) { GLuint fbo = 0; @@ -1812,9 +1851,7 @@ void EffectChain::execute_phase(Phase *phase, bool last_phase, mapinsert(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. @@ -1845,7 +1882,7 @@ void EffectChain::execute_phase(Phase *phase, bool last_phase, mapeffects.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()) { @@ -1860,12 +1897,34 @@ void EffectChain::execute_phase(Phase *phase, bool last_phase, map::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::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();