X-Git-Url: https://git.sesse.net/?p=movit;a=blobdiff_plain;f=effect_chain.cpp;h=b6b5f59925df18b83ea75100e276e6d153076310;hp=519cb2c6b9ea1c5816ead123b05260904519f96e;hb=236a0ad8b604d5b3bff53f40b600991168f76800;hpb=b564238fa1293c01c77bcabe7b2de267f146ab24 diff --git a/effect_chain.cpp b/effect_chain.cpp index 519cb2c..b6b5f59 100644 --- a/effect_chain.cpp +++ b/effect_chain.cpp @@ -12,6 +12,7 @@ #include #include #include +#include #include "alpha_division_effect.h" #include "alpha_multiplication_effect.h" @@ -19,13 +20,16 @@ #include "dither_effect.h" #include "effect.h" #include "effect_chain.h" +#include "effect_util.h" #include "gamma_compression_effect.h" #include "gamma_expansion_effect.h" #include "init.h" #include "input.h" #include "resource_pool.h" #include "util.h" +#include "ycbcr_conversion_effect.h" +using namespace Eigen; using namespace std; namespace movit { @@ -33,8 +37,11 @@ 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) { @@ -72,8 +79,25 @@ 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_rgba = true; +} + +void EffectChain::add_ycbcr_output(const ImageFormat &format, OutputAlphaFormat alpha_format, + const YCbCrFormat &ycbcr_format, YCbCrOutputSplitting output_splitting) +{ + assert(!finalized); + assert(!output_color_ycbcr); + output_format = format; + output_alpha_format = alpha_format; + 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); } Node *EffectChain::add_node(Effect *effect) @@ -188,7 +212,7 @@ Effect *EffectChain::add_effect(Effect *effect, const vector &inputs) return effect; } -// GLSL pre-1.30 doesn't support token pasting. Replace PREFIX(x) with _x. +// ESSL doesn't support token pasting. Replace PREFIX(x) with _x. string replace_prefix(const string &text, const string &prefix) { string output; @@ -229,11 +253,61 @@ string replace_prefix(const string &text, const string &prefix) return output; } +namespace { + +template +void extract_uniform_declarations(const vector > &effect_uniforms, + const string &type_specifier, + const string &effect_id, + vector > *phase_uniforms, + string *glsl_string) +{ + for (unsigned i = 0; i < effect_uniforms.size(); ++i) { + phase_uniforms->push_back(effect_uniforms[i]); + phase_uniforms->back().prefix = effect_id; + + *glsl_string += string("uniform ") + type_specifier + " " + effect_id + + "_" + effect_uniforms[i].name + ";\n"; + } +} + +template +void extract_uniform_array_declarations(const vector > &effect_uniforms, + const string &type_specifier, + const string &effect_id, + vector > *phase_uniforms, + string *glsl_string) +{ + for (unsigned i = 0; i < effect_uniforms.size(); ++i) { + phase_uniforms->push_back(effect_uniforms[i]); + phase_uniforms->back().prefix = effect_id; + + char buf[256]; + snprintf(buf, sizeof(buf), "uniform %s %s_%s[%d];\n", + type_specifier.c_str(), effect_id.c_str(), + effect_uniforms[i].name.c_str(), + int(effect_uniforms[i].num_values)); + *glsl_string += buf; + } +} + +template +void collect_uniform_locations(GLuint glsl_program_num, vector > *phase_uniforms) +{ + for (unsigned i = 0; i < phase_uniforms->size(); ++i) { + Uniform &uniform = (*phase_uniforms)[i]; + uniform.location = get_uniform_location(glsl_program_num, uniform.prefix, uniform.name); + } +} + +} // namespace + void EffectChain::compile_glsl_program(Phase *phase) { - string frag_shader = read_version_dependent_file("header", "frag"); + string frag_shader_header = read_version_dependent_file("header", "frag"); + string frag_shader = ""; - // Create functions for all the texture inputs that we need. + // Create functions and uniforms for all the texture inputs that we need. for (unsigned i = 0; i < phase->inputs.size(); ++i) { Node *input = phase->inputs[i]->output_node; char effect_id[256]; @@ -245,14 +319,27 @@ void EffectChain::compile_glsl_program(Phase *phase) frag_shader += "\treturn tex2D(tex_" + string(effect_id) + ", tc);\n"; frag_shader += "}\n"; frag_shader += "\n"; + + Uniform uniform; + uniform.name = effect_id; + uniform.value = &phase->input_samplers[i]; + uniform.prefix = "tex"; + uniform.num_values = 1; + uniform.location = -1; + phase->uniforms_sampler2d.push_back(uniform); } + // Give each effect in the phase its own ID. for (unsigned i = 0; i < phase->effects.size(); ++i) { Node *node = phase->effects[i]; char effect_id[256]; sprintf(effect_id, "eff%u", i); phase->effect_ids.insert(make_pair(node, effect_id)); + } + for (unsigned i = 0; i < phase->effects.size(); ++i) { + Node *node = phase->effects[i]; + const string effect_id = phase->effect_ids[node]; if (node->incoming_links.size() == 1) { frag_shader += string("#define INPUT ") + phase->effect_ids[node->incoming_links[0]] + "\n"; } else { @@ -265,7 +352,6 @@ void EffectChain::compile_glsl_program(Phase *phase) frag_shader += "\n"; frag_shader += string("#define FUNCNAME ") + effect_id + "\n"; - frag_shader += replace_prefix(node->effect->output_convenience_uniforms(), effect_id); frag_shader += replace_prefix(node->effect->output_fragment_shader(), effect_id); frag_shader += "#undef PREFIX\n"; frag_shader += "#undef FUNCNAME\n"; @@ -281,10 +367,80 @@ 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. + if (phase->output_node->outgoing_links.empty() && output_color_ycbcr) { + switch (output_ycbcr_splitting) { + case YCBCR_OUTPUT_INTERLEAVED: + // No #defines set. + break; + case YCBCR_OUTPUT_SPLIT_Y_AND_CBCR: + frag_shader += "#define YCBCR_OUTPUT_SPLIT_Y_AND_CBCR 1\n"; + break; + case YCBCR_OUTPUT_PLANAR: + frag_shader += "#define YCBCR_OUTPUT_PLANAR 1\n"; + 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.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 + // before in the output source, since output_fragment_shader() is allowed + // to register new uniforms (e.g. arrays that are of unknown length until + // finalization time). + // TODO: Make a uniform block for platforms that support it. + string frag_shader_uniforms = ""; + for (unsigned i = 0; i < phase->effects.size(); ++i) { + Node *node = phase->effects[i]; + Effect *effect = node->effect; + const string effect_id = phase->effect_ids[node]; + extract_uniform_declarations(effect->uniforms_sampler2d, "sampler2D", effect_id, &phase->uniforms_sampler2d, &frag_shader_uniforms); + extract_uniform_declarations(effect->uniforms_bool, "bool", effect_id, &phase->uniforms_bool, &frag_shader_uniforms); + extract_uniform_declarations(effect->uniforms_int, "int", effect_id, &phase->uniforms_int, &frag_shader_uniforms); + extract_uniform_declarations(effect->uniforms_float, "float", effect_id, &phase->uniforms_float, &frag_shader_uniforms); + 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_vec2_array, "vec2", effect_id, &phase->uniforms_vec2, &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); + } + + frag_shader = frag_shader_header + frag_shader_uniforms + frag_shader; string vert_shader = read_version_dependent_file("vs", "vert"); + + // 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); + + // Collect the resulting location numbers for each uniform. + collect_uniform_locations(phase->glsl_program_num, &phase->uniforms_sampler2d); + collect_uniform_locations(phase->glsl_program_num, &phase->uniforms_bool); + collect_uniform_locations(phase->glsl_program_num, &phase->uniforms_int); + collect_uniform_locations(phase->glsl_program_num, &phase->uniforms_float); + collect_uniform_locations(phase->glsl_program_num, &phase->uniforms_vec2); + collect_uniform_locations(phase->glsl_program_num, &phase->uniforms_vec3); + collect_uniform_locations(phase->glsl_program_num, &phase->uniforms_vec4); + collect_uniform_locations(phase->glsl_program_num, &phase->uniforms_mat3); } // Construct GLSL programs, starting at the given effect and following @@ -419,6 +575,9 @@ Phase *EffectChain::construct_phase(Node *output, map *complete sort(phase->inputs.begin(), phase->inputs.end()); phase->inputs.erase(unique(phase->inputs.begin(), phase->inputs.end()), phase->inputs.end()); + // Allocate samplers for each input. + phase->input_samplers.resize(phase->inputs.size()); + // We added the effects from the output and back, but we need to output // them in topological sort order in the shader. phase->effects = topological_sort(phase->effects); @@ -1363,6 +1522,22 @@ void EffectChain::fix_output_gamma() connect_nodes(output, conversion); } } + +// If the user has requested Y'CbCr output, we need to do this conversion +// _after_ GammaCompressionEffect etc., but before dither (see below). +// This is because Y'CbCr, with the exception of a special optional mode +// in Rec. 2020 (which we currently don't support), is defined to work on +// gamma-encoded data. +void EffectChain::add_ycbcr_conversion_if_needed() +{ + assert(output_color_rgba || output_color_ycbcr); + if (!output_color_ycbcr) { + return; + } + Node *output = find_output_node(); + Node *ycbcr = add_node(new YCbCrConversionEffect(output_ycbcr_format)); + connect_nodes(output, ycbcr); +} // If the user has requested dither, add a DitherEffect right at the end // (after GammaCompressionEffect etc.). This needs to be done after everything else, @@ -1445,11 +1620,13 @@ void EffectChain::finalize() fix_internal_gamma_by_asking_inputs(15); fix_internal_gamma_by_inserting_nodes(16); - output_dot("step17-before-dither.dot"); + output_dot("step17-before-ycbcr.dot"); + add_ycbcr_conversion_if_needed(); + output_dot("step18-before-dither.dot"); add_dither_if_needed(); - output_dot("step18-final.dot"); + output_dot("step19-final.dot"); // Construct all needed GLSL programs, starting at the output. // We need to keep track of which effects have already been computed, @@ -1458,7 +1635,7 @@ void EffectChain::finalize() map completed_effects; construct_phase(find_output_node(), &completed_effects); - output_dot("step19-split-to-phases.dot"); + output_dot("step20-split-to-phases.dot"); assert(phases[0]->inputs.empty()); @@ -1469,6 +1646,10 @@ 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. + glDisable(GL_DITHER); + // Save original viewport. GLuint x = 0, y = 0; @@ -1489,6 +1670,23 @@ 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 + }; + + 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. + set generated_mipmaps; // We choose the simplest option of having one texture per output, @@ -1530,6 +1728,12 @@ 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); + + glDeleteVertexArrays(1, &vao); + check_error(); + if (do_phase_timing) { // Get back the timer queries. for (unsigned phase_num = 0; phase_num < phases.size(); ++phase_num) { @@ -1612,7 +1816,8 @@ void EffectChain::execute_phase(Phase *phase, bool last_phase, mapinsert(input); } - setup_rtt_sampler(glsl_program_num, sampler, phase->effect_ids[input->output_node], phase->input_needs_mipmaps); + setup_rtt_sampler(sampler, phase->input_needs_mipmaps); + phase->input_samplers[sampler] = sampler; // Bind the sampler to the right uniform. } // And now the output. (Already set up for us if it is the last phase.) @@ -1638,28 +1843,13 @@ void EffectChain::execute_phase(Phase *phase, bool last_phase, maprelease_fbo(fbo); } +} - glDeleteVertexArrays(1, &vao); - check_error(); +void EffectChain::setup_uniforms(Phase *phase) +{ + // TODO: Use UBO blocks. + for (size_t i = 0; i < phase->uniforms_sampler2d.size(); ++i) { + const Uniform &uniform = phase->uniforms_sampler2d[i]; + if (uniform.location != -1) { + glUniform1iv(uniform.location, uniform.num_values, uniform.value); + } + } + for (size_t i = 0; i < phase->uniforms_bool.size(); ++i) { + const Uniform &uniform = phase->uniforms_bool[i]; + assert(uniform.num_values == 1); + if (uniform.location != -1) { + glUniform1i(uniform.location, *uniform.value); + } + } + for (size_t i = 0; i < phase->uniforms_int.size(); ++i) { + const Uniform &uniform = phase->uniforms_int[i]; + if (uniform.location != -1) { + glUniform1iv(uniform.location, uniform.num_values, uniform.value); + } + } + for (size_t i = 0; i < phase->uniforms_float.size(); ++i) { + const Uniform &uniform = phase->uniforms_float[i]; + if (uniform.location != -1) { + glUniform1fv(uniform.location, uniform.num_values, uniform.value); + } + } + for (size_t i = 0; i < phase->uniforms_vec2.size(); ++i) { + const Uniform &uniform = phase->uniforms_vec2[i]; + if (uniform.location != -1) { + glUniform2fv(uniform.location, uniform.num_values, uniform.value); + } + } + for (size_t i = 0; i < phase->uniforms_vec3.size(); ++i) { + const Uniform &uniform = phase->uniforms_vec3[i]; + if (uniform.location != -1) { + glUniform3fv(uniform.location, uniform.num_values, uniform.value); + } + } + for (size_t i = 0; i < phase->uniforms_vec4.size(); ++i) { + const Uniform &uniform = phase->uniforms_vec4[i]; + if (uniform.location != -1) { + glUniform4fv(uniform.location, uniform.num_values, uniform.value); + } + } + for (size_t i = 0; i < phase->uniforms_mat3.size(); ++i) { + const Uniform &uniform = phase->uniforms_mat3[i]; + assert(uniform.num_values == 1); + if (uniform.location != -1) { + // Convert to float (GLSL has no double matrices). + float matrixf[9]; + for (unsigned y = 0; y < 3; ++y) { + for (unsigned x = 0; x < 3; ++x) { + matrixf[y + x * 3] = (*uniform.value)(y, x); + } + } + glUniformMatrix3fv(uniform.location, 1, GL_FALSE, matrixf); + } + } } -void EffectChain::setup_rtt_sampler(GLuint glsl_program_num, int sampler_num, const string &effect_id, bool use_mipmaps) +void EffectChain::setup_rtt_sampler(int sampler_num, bool use_mipmaps) { glActiveTexture(GL_TEXTURE0 + sampler_num); check_error(); @@ -1691,10 +1940,6 @@ void EffectChain::setup_rtt_sampler(GLuint glsl_program_num, int sampler_num, co check_error(); glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE); check_error(); - - string texture_name = string("tex_") + effect_id; - glUniform1i(glGetUniformLocation(glsl_program_num, texture_name.c_str()), sampler_num); - check_error(); } } // namespace movit