X-Git-Url: https://git.sesse.net/?p=movit;a=blobdiff_plain;f=effect_chain.cpp;h=d80c4f82a28b33c39c983b38f02894ad514979e1;hp=a9b892725358039173725feea5306692f3432988;hb=85f9719bf3519b1f1942738d11601584f5d38725;hpb=7221906173f1cf1ce6913cbe9d62d5ca11e9ee0d diff --git a/effect_chain.cpp b/effect_chain.cpp index a9b8927..d80c4f8 100644 --- a/effect_chain.cpp +++ b/effect_chain.cpp @@ -1,57 +1,66 @@ #define GL_GLEXT_PROTOTYPES 1 -#include +#include +#include +#include #include +#include +#include +#include #include -#include -#include - #include #include #include #include -#include "util.h" -#include "effect_chain.h" -#include "gamma_expansion_effect.h" -#include "gamma_compression_effect.h" -#include "colorspace_conversion_effect.h" -#include "alpha_multiplication_effect.h" #include "alpha_division_effect.h" +#include "alpha_multiplication_effect.h" +#include "colorspace_conversion_effect.h" #include "dither_effect.h" -#include "input.h" +#include "effect.h" +#include "effect_chain.h" +#include "gamma_compression_effect.h" +#include "gamma_expansion_effect.h" #include "init.h" +#include "input.h" +#include "resource_pool.h" +#include "util.h" -EffectChain::EffectChain(float aspect_nom, float aspect_denom) +using namespace std; + +EffectChain::EffectChain(float aspect_nom, float aspect_denom, ResourcePool *resource_pool) : aspect_nom(aspect_nom), aspect_denom(aspect_denom), dither_effect(NULL), - fbo(0), num_dither_bits(0), - finalized(false) {} + finalized(false), + resource_pool(resource_pool) { + if (resource_pool == NULL) { + this->resource_pool = new ResourcePool(); + owns_resource_pool = true; + } else { + owns_resource_pool = false; + } +} EffectChain::~EffectChain() { for (unsigned i = 0; i < nodes.size(); ++i) { - if (nodes[i]->output_texture != 0) { - glDeleteTextures(1, &nodes[i]->output_texture); - } delete nodes[i]->effect; delete nodes[i]; } for (unsigned i = 0; i < phases.size(); ++i) { - glDeleteProgram(phases[i]->glsl_program_num); - glDeleteShader(phases[i]->vertex_shader); - glDeleteShader(phases[i]->fragment_shader); + resource_pool->release_glsl_program(phases[i]->glsl_program_num); delete phases[i]; } - if (fbo != 0) { - glDeleteFramebuffers(1, &fbo); + if (owns_resource_pool) { + delete resource_pool; } } Input *EffectChain::add_input(Input *input) { + assert(!finalized); inputs.push_back(input); add_node(input); return input; @@ -59,26 +68,27 @@ Input *EffectChain::add_input(Input *input) void EffectChain::add_output(const ImageFormat &format, OutputAlphaFormat alpha_format) { + assert(!finalized); output_format = format; output_alpha_format = alpha_format; } Node *EffectChain::add_node(Effect *effect) { - char effect_id[256]; - sprintf(effect_id, "eff%u", (unsigned)nodes.size()); + for (unsigned i = 0; i < nodes.size(); ++i) { + assert(nodes[i]->effect != effect); + } Node *node = new Node; node->effect = effect; node->disabled = false; - node->effect_id = effect_id; node->output_color_space = COLORSPACE_INVALID; node->output_gamma_curve = GAMMA_INVALID; node->output_alpha_type = ALPHA_INVALID; - node->output_texture = 0; nodes.push_back(node); node_map[effect] = node; + effect->inform_added(this); return node; } @@ -136,7 +146,7 @@ void EffectChain::insert_node_between(Node *sender, Node *middle, Node *receiver assert(middle->incoming_links.size() == middle->effect->num_inputs()); } -void EffectChain::find_all_nonlinear_inputs(Node *node, std::vector *nonlinear_inputs) +void EffectChain::find_all_nonlinear_inputs(Node *node, vector *nonlinear_inputs) { if (node->output_gamma_curve == GAMMA_LINEAR && node->effect->effect_type_id() != "GammaCompressionEffect") { @@ -152,8 +162,9 @@ void EffectChain::find_all_nonlinear_inputs(Node *node, std::vector *non } } -Effect *EffectChain::add_effect(Effect *effect, const std::vector &inputs) +Effect *EffectChain::add_effect(Effect *effect, const vector &inputs) { + assert(!finalized); assert(inputs.size() == effect->num_inputs()); Node *node = add_node(effect); for (unsigned i = 0; i < inputs.size(); ++i) { @@ -164,15 +175,15 @@ Effect *EffectChain::add_effect(Effect *effect, const std::vector &inp } // GLSL pre-1.30 doesn't support token pasting. Replace PREFIX(x) with _x. -std::string replace_prefix(const std::string &text, const std::string &prefix) +string replace_prefix(const string &text, const string &prefix) { - std::string output; + string output; size_t start = 0; while (start < text.size()) { size_t pos = text.find("PREFIX(", start); - if (pos == std::string::npos) { - output.append(text.substr(start, std::string::npos)); + if (pos == string::npos) { + output.append(text.substr(start, string::npos)); break; } @@ -205,49 +216,56 @@ std::string replace_prefix(const std::string &text, const std::string &prefix) } Phase *EffectChain::compile_glsl_program( - const std::vector &inputs, - const std::vector &effects) + const vector &inputs, + const vector &effects) { + Phase *phase = new Phase; assert(!effects.empty()); // Deduplicate the inputs. - std::vector true_inputs = inputs; - std::sort(true_inputs.begin(), true_inputs.end()); - true_inputs.erase(std::unique(true_inputs.begin(), true_inputs.end()), true_inputs.end()); + vector true_inputs = inputs; + sort(true_inputs.begin(), true_inputs.end()); + true_inputs.erase(unique(true_inputs.begin(), true_inputs.end()), true_inputs.end()); bool input_needs_mipmaps = false; - std::string frag_shader = read_file("header.frag"); + string frag_shader = read_file("header.frag"); // Create functions for all the texture inputs that we need. for (unsigned i = 0; i < true_inputs.size(); ++i) { Node *input = true_inputs[i]; + char effect_id[256]; + sprintf(effect_id, "in%u", i); + phase->effect_ids.insert(make_pair(input, effect_id)); - frag_shader += std::string("uniform sampler2D tex_") + input->effect_id + ";\n"; - frag_shader += std::string("vec4 ") + input->effect_id + "(vec2 tc) {\n"; - frag_shader += "\treturn texture2D(tex_" + input->effect_id + ", tc);\n"; + frag_shader += string("uniform sampler2D tex_") + effect_id + ";\n"; + frag_shader += string("vec4 ") + effect_id + "(vec2 tc) {\n"; + frag_shader += "\treturn texture2D(tex_" + string(effect_id) + ", tc);\n"; frag_shader += "}\n"; frag_shader += "\n"; } - std::vector sorted_effects = topological_sort(effects); + vector sorted_effects = topological_sort(effects); for (unsigned i = 0; i < sorted_effects.size(); ++i) { Node *node = sorted_effects[i]; + char effect_id[256]; + sprintf(effect_id, "eff%u", i); + phase->effect_ids.insert(make_pair(node, effect_id)); if (node->incoming_links.size() == 1) { - frag_shader += std::string("#define INPUT ") + node->incoming_links[0]->effect_id + "\n"; + frag_shader += string("#define INPUT ") + phase->effect_ids[node->incoming_links[0]] + "\n"; } else { for (unsigned j = 0; j < node->incoming_links.size(); ++j) { char buf[256]; - sprintf(buf, "#define INPUT%d %s\n", j + 1, node->incoming_links[j]->effect_id.c_str()); + sprintf(buf, "#define INPUT%d %s\n", j + 1, phase->effect_ids[node->incoming_links[j]].c_str()); frag_shader += buf; } } frag_shader += "\n"; - frag_shader += std::string("#define FUNCNAME ") + node->effect_id + "\n"; - frag_shader += replace_prefix(node->effect->output_convenience_uniforms(), node->effect_id); - frag_shader += replace_prefix(node->effect->output_fragment_shader(), node->effect_id); + 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"; if (node->incoming_links.size() == 1) { @@ -269,37 +287,10 @@ Phase *EffectChain::compile_glsl_program( CHECK(node->effect->set_int("needs_mipmaps", input_needs_mipmaps)); } } - frag_shader += std::string("#define INPUT ") + sorted_effects.back()->effect_id + "\n"; + frag_shader += string("#define INPUT ") + phase->effect_ids[sorted_effects.back()] + "\n"; frag_shader.append(read_file("footer.frag")); - if (movit_debug_level == MOVIT_DEBUG_ON) { - // Output shader to a temporary file, for easier debugging. - static int compiled_shader_num = 0; - char filename[256]; - sprintf(filename, "chain-%03d.frag", compiled_shader_num++); - FILE *fp = fopen(filename, "w"); - if (fp == NULL) { - perror(filename); - exit(1); - } - fprintf(fp, "%s\n", frag_shader.c_str()); - fclose(fp); - } - - GLuint glsl_program_num = glCreateProgram(); - GLuint vs_obj = compile_shader(read_file("vs.vert"), GL_VERTEX_SHADER); - GLuint fs_obj = compile_shader(frag_shader, GL_FRAGMENT_SHADER); - glAttachShader(glsl_program_num, vs_obj); - check_error(); - glAttachShader(glsl_program_num, fs_obj); - check_error(); - glLinkProgram(glsl_program_num); - check_error(); - - Phase *phase = new Phase; - phase->glsl_program_num = glsl_program_num; - phase->vertex_shader = vs_obj; - phase->fragment_shader = fs_obj; + phase->glsl_program_num = resource_pool->compile_glsl_program(read_file("vs.vert"), frag_shader); phase->input_needs_mipmaps = input_needs_mipmaps; phase->inputs = true_inputs; phase->effects = sorted_effects; @@ -317,25 +308,25 @@ Phase *EffectChain::compile_glsl_program( // without any explicit recursion. void EffectChain::construct_glsl_programs(Node *output) { - // Which effects have already been completed in this phase? + // Which effects have already been completed? // We need to keep track of it, as an effect with multiple outputs // could otherwise be calculated multiple times. - std::set completed_effects; + set completed_effects; // Effects in the current phase, as well as inputs (outputs from other phases // that we depend on). Note that since we start iterating from the end, // the effect list will be in the reverse order. - std::vector this_phase_inputs; - std::vector this_phase_effects; + vector this_phase_inputs; + vector this_phase_effects; // Effects that we have yet to calculate, but that we know should // be in the current phase. - std::stack effects_todo_this_phase; + stack effects_todo_this_phase; // Effects that we have yet to calculate, but that come from other phases. // We delay these until we have this phase done in its entirety, // at which point we pick any of them and start a new phase from that. - std::stack effects_todo_other_phases; + stack effects_todo_other_phases; effects_todo_this_phase.push(output); @@ -348,16 +339,19 @@ void EffectChain::construct_glsl_programs(Node *output) // This should currently only happen for effects that are inputs // (either true inputs or phase outputs). We special-case inputs, // and then deduplicate phase outputs in compile_glsl_program(). - if (node->effect->num_inputs() == 0 && completed_effects.count(node)) { - continue; + if (node->effect->num_inputs() == 0) { + if (find(this_phase_effects.begin(), this_phase_effects.end(), node) != this_phase_effects.end()) { + continue; + } + } else { + assert(completed_effects.count(node) == 0); } - assert(completed_effects.count(node) == 0); this_phase_effects.push_back(node); completed_effects.insert(node); // Find all the dependencies of this effect, and add them to the stack. - std::vector deps = node->incoming_links; + vector deps = node->incoming_links; assert(node->effect->num_inputs() == deps.size()); for (unsigned i = 0; i < deps.size(); ++i) { bool start_new_phase = false; @@ -432,7 +426,7 @@ void EffectChain::construct_glsl_programs(Node *output) // Finally, since the phases are found from the output but must be executed // from the input(s), reverse them, too. - std::reverse(phases.begin(), phases.end()); + reverse(phases.begin(), phases.end()); } void EffectChain::output_dot(const char *filename) @@ -451,21 +445,26 @@ void EffectChain::output_dot(const char *filename) fprintf(fp, " output [shape=box label=\"(output)\"];\n"); for (unsigned i = 0; i < nodes.size(); ++i) { // Find out which phase this event belongs to. - int in_phase = -1; + vector in_phases; for (unsigned j = 0; j < phases.size(); ++j) { const Phase* p = phases[j]; - if (std::find(p->effects.begin(), p->effects.end(), nodes[i]) != p->effects.end()) { - assert(in_phase == -1); - in_phase = j; + if (find(p->effects.begin(), p->effects.end(), nodes[i]) != p->effects.end()) { + in_phases.push_back(j); } } - if (in_phase == -1) { + if (in_phases.empty()) { fprintf(fp, " n%ld [label=\"%s\"];\n", (long)nodes[i], nodes[i]->effect->effect_type_id().c_str()); - } else { + } else if (in_phases.size() == 1) { fprintf(fp, " n%ld [label=\"%s\" style=\"filled\" fillcolor=\"/accent8/%d\"];\n", (long)nodes[i], nodes[i]->effect->effect_type_id().c_str(), - (in_phase % 8) + 1); + (in_phases[0] % 8) + 1); + } else { + // If we had new enough Graphviz, style="wedged" would probably be ideal here. + // But alas. + fprintf(fp, " n%ld [label=\"%s [in multiple phases]\" style=\"filled\" fillcolor=\"/accent8/%d\"];\n", + (long)nodes[i], nodes[i]->effect->effect_type_id().c_str(), + (in_phases[0] % 8) + 1); } char from_node_id[256]; @@ -475,13 +474,13 @@ void EffectChain::output_dot(const char *filename) char to_node_id[256]; snprintf(to_node_id, 256, "n%ld", (long)nodes[i]->outgoing_links[j]); - std::vector labels = get_labels_for_edge(nodes[i], nodes[i]->outgoing_links[j]); + vector labels = get_labels_for_edge(nodes[i], nodes[i]->outgoing_links[j]); output_dot_edge(fp, from_node_id, to_node_id, labels); } if (nodes[i]->outgoing_links.empty() && !nodes[i]->disabled) { // Output node. - std::vector labels = get_labels_for_edge(nodes[i], NULL); + vector labels = get_labels_for_edge(nodes[i], NULL); output_dot_edge(fp, from_node_id, "output", labels); } } @@ -490,9 +489,9 @@ void EffectChain::output_dot(const char *filename) fclose(fp); } -std::vector EffectChain::get_labels_for_edge(const Node *from, const Node *to) +vector EffectChain::get_labels_for_edge(const Node *from, const Node *to) { - std::vector labels; + vector labels; if (to != NULL && to->effect->needs_texture_bounce()) { labels.push_back("needs_bounce"); @@ -547,14 +546,14 @@ std::vector EffectChain::get_labels_for_edge(const Node *from, cons } void EffectChain::output_dot_edge(FILE *fp, - const std::string &from_node_id, - const std::string &to_node_id, - const std::vector &labels) + const string &from_node_id, + const string &to_node_id, + const vector &labels) { if (labels.empty()) { fprintf(fp, " %s -> %s;\n", from_node_id.c_str(), to_node_id.c_str()); } else { - std::string label = labels[0]; + string label = labels[0]; for (unsigned k = 1; k < labels.size(); ++k) { label += ", " + labels[k]; } @@ -562,16 +561,30 @@ void EffectChain::output_dot_edge(FILE *fp, } } -unsigned EffectChain::fit_rectangle_to_aspect(unsigned width, unsigned height) +void EffectChain::size_rectangle_to_fit(unsigned width, unsigned height, unsigned *output_width, unsigned *output_height) { + unsigned scaled_width, scaled_height; + if (float(width) * aspect_denom >= float(height) * aspect_nom) { // Same aspect, or W/H > aspect (image is wider than the frame). - // In either case, keep width. - return width; + // In either case, keep width, and adjust height. + scaled_width = width; + scaled_height = lrintf(width * aspect_denom / aspect_nom); } else { // W/H < aspect (image is taller than the frame), so keep height, - // and adjust width correspondingly. - return lrintf(height * aspect_nom / aspect_denom); + // and adjust width. + scaled_width = lrintf(height * aspect_nom / aspect_denom); + scaled_height = height; + } + + // We should be consistently larger or smaller then the existing choice, + // since we have the same aspect. + assert(!(scaled_width < *output_width && scaled_height > *output_height)); + assert(!(scaled_height < *output_height && scaled_width > *output_width)); + + if (scaled_width >= *output_width && scaled_height >= *output_height) { + *output_width = scaled_width; + *output_height = scaled_height; } } @@ -595,8 +608,8 @@ void EffectChain::inform_input_sizes(Phase *phase) } for (unsigned i = 0; i < phase->inputs.size(); ++i) { Node *input = phase->inputs[i]; - input->output_width = input->phase->output_width; - input->output_height = input->phase->output_height; + input->output_width = input->phase->virtual_output_width; + input->output_height = input->phase->virtual_output_height; assert(input->output_width != 0); assert(input->output_height != 0); } @@ -639,20 +652,25 @@ void EffectChain::find_output_size(Phase *phase) // If the last effect explicitly sets an output size, use that. if (output_node->effect->changes_output_size()) { - output_node->effect->get_output_size(&phase->output_width, &phase->output_height); + output_node->effect->get_output_size(&phase->output_width, &phase->output_height, + &phase->virtual_output_width, &phase->virtual_output_height); return; } - // If not, look at the input phases and textures. - // We select the largest one (by fit into the current aspect). - unsigned best_width = 0; + // If all effects have the same size, use that. + unsigned output_width = 0, output_height = 0; + bool all_inputs_same_size = true; + for (unsigned i = 0; i < phase->inputs.size(); ++i) { Node *input = phase->inputs[i]; assert(input->phase->output_width != 0); assert(input->phase->output_height != 0); - unsigned width = fit_rectangle_to_aspect(input->phase->output_width, input->phase->output_height); - if (width > best_width) { - best_width = width; + if (output_width == 0 && output_height == 0) { + output_width = input->phase->virtual_output_width; + output_height = input->phase->virtual_output_height; + } else if (output_width != input->phase->virtual_output_width || + output_height != input->phase->virtual_output_height) { + all_inputs_same_size = false; } } for (unsigned i = 0; i < phase->effects.size(); ++i) { @@ -662,14 +680,45 @@ void EffectChain::find_output_size(Phase *phase) } Input *input = static_cast(effect); - unsigned width = fit_rectangle_to_aspect(input->get_width(), input->get_height()); - if (width > best_width) { - best_width = width; + if (output_width == 0 && output_height == 0) { + output_width = input->get_width(); + output_height = input->get_height(); + } else if (output_width != input->get_width() || + output_height != input->get_height()) { + all_inputs_same_size = false; + } + } + + if (all_inputs_same_size) { + assert(output_width != 0); + assert(output_height != 0); + phase->virtual_output_width = phase->output_width = output_width; + phase->virtual_output_height = phase->output_height = output_height; + return; + } + + // If not, fit all the inputs into the current aspect, and select the largest one. + output_width = 0; + output_height = 0; + for (unsigned i = 0; i < phase->inputs.size(); ++i) { + Node *input = phase->inputs[i]; + assert(input->phase->output_width != 0); + assert(input->phase->output_height != 0); + size_rectangle_to_fit(input->phase->output_width, input->phase->output_height, &output_width, &output_height); + } + for (unsigned i = 0; i < phase->effects.size(); ++i) { + Effect *effect = phase->effects[i]->effect; + if (effect->num_inputs() != 0) { + continue; } + + Input *input = static_cast(effect); + size_rectangle_to_fit(input->get_width(), input->get_height(), &output_width, &output_height); } - assert(best_width != 0); - phase->output_width = best_width; - phase->output_height = best_width * aspect_denom / aspect_nom; + assert(output_width != 0); + assert(output_height != 0); + phase->virtual_output_width = phase->output_width = output_width; + phase->virtual_output_height = phase->output_height = output_height; } void EffectChain::sort_all_nodes_topologically() @@ -677,10 +726,10 @@ void EffectChain::sort_all_nodes_topologically() nodes = topological_sort(nodes); } -std::vector EffectChain::topological_sort(const std::vector &nodes) +vector EffectChain::topological_sort(const vector &nodes) { - std::set nodes_left_to_visit(nodes.begin(), nodes.end()); - std::vector sorted_list; + set nodes_left_to_visit(nodes.begin(), nodes.end()); + vector sorted_list; for (unsigned i = 0; i < nodes.size(); ++i) { topological_sort_visit_node(nodes[i], &nodes_left_to_visit, &sorted_list); } @@ -688,7 +737,7 @@ std::vector EffectChain::topological_sort(const std::vector &nod return sorted_list; } -void EffectChain::topological_sort_visit_node(Node *node, std::set *nodes_left_to_visit, std::vector *sorted_list) +void EffectChain::topological_sort_visit_node(Node *node, set *nodes_left_to_visit, vector *sorted_list) { if (nodes_left_to_visit->count(node) == 0) { return; @@ -717,12 +766,13 @@ void EffectChain::find_color_spaces_for_inputs() case Effect::OUTPUT_BLANK_ALPHA: node->output_alpha_type = ALPHA_BLANK; break; - case Effect::INPUT_AND_OUTPUT_ALPHA_PREMULTIPLIED: + case Effect::INPUT_AND_OUTPUT_PREMULTIPLIED_ALPHA: node->output_alpha_type = ALPHA_PREMULTIPLIED; break; - case Effect::OUTPUT_ALPHA_POSTMULTIPLIED: + case Effect::OUTPUT_POSTMULTIPLIED_ALPHA: node->output_alpha_type = ALPHA_POSTMULTIPLIED; break; + case Effect::INPUT_PREMULTIPLIED_ALPHA_KEEP_BLANK: case Effect::DONT_CARE_ALPHA_TYPE: default: assert(false); @@ -826,7 +876,7 @@ void EffectChain::propagate_alpha() } // Only inputs can have unconditional alpha output (OUTPUT_BLANK_ALPHA - // or OUTPUT_ALPHA_POSTMULTIPLIED), and they have already been + // or OUTPUT_POSTMULTIPLIED_ALPHA), and they have already been // taken care of above. Rationale: Even if you could imagine // e.g. an effect that took in an image and set alpha=1.0 // unconditionally, it wouldn't make any sense to have it as @@ -834,7 +884,8 @@ void EffectChain::propagate_alpha() // got its input pre- or postmultiplied, so it wouldn't know // whether to divide away the old alpha or not. Effect::AlphaHandling alpha_handling = node->effect->alpha_handling(); - assert(alpha_handling == Effect::INPUT_AND_OUTPUT_ALPHA_PREMULTIPLIED || + assert(alpha_handling == Effect::INPUT_AND_OUTPUT_PREMULTIPLIED_ALPHA || + alpha_handling == Effect::INPUT_PREMULTIPLIED_ALPHA_KEEP_BLANK || alpha_handling == Effect::DONT_CARE_ALPHA_TYPE); // If the node has multiple inputs, check that they are all valid and @@ -874,16 +925,16 @@ void EffectChain::propagate_alpha() continue; } - if (alpha_handling == Effect::INPUT_AND_OUTPUT_ALPHA_PREMULTIPLIED) { + if (alpha_handling == Effect::INPUT_AND_OUTPUT_PREMULTIPLIED_ALPHA || + alpha_handling == Effect::INPUT_PREMULTIPLIED_ALPHA_KEEP_BLANK) { // If the effect has asked for premultiplied alpha, check that it has got it. if (any_postmultiplied) { node->output_alpha_type = ALPHA_INVALID; + } else if (!any_premultiplied && + alpha_handling == Effect::INPUT_PREMULTIPLIED_ALPHA_KEEP_BLANK) { + // Blank input alpha, and the effect preserves blank alpha. + node->output_alpha_type = ALPHA_BLANK; } else { - // In some rare cases, it might be advantageous to say - // that blank input alpha yields blank output alpha. - // However, this would cause a more complex Effect interface - // an effect would need to guarantee that it doesn't mess with - // blank alpha), so this is the simplest. node->output_alpha_type = ALPHA_PREMULTIPLIED; } } else { @@ -938,7 +989,7 @@ void EffectChain::fix_internal_color_spaces() } // Go through each input that is not sRGB, and insert - // a colorspace conversion before it. + // a colorspace conversion after it. for (unsigned j = 0; j < node->incoming_links.size(); ++j) { Node *input = node->incoming_links[j]; assert(input->output_color_space != COLORSPACE_INVALID); @@ -949,7 +1000,8 @@ void EffectChain::fix_internal_color_spaces() CHECK(conversion->effect->set_int("source_space", input->output_color_space)); CHECK(conversion->effect->set_int("destination_space", COLORSPACE_sRGB)); conversion->output_color_space = COLORSPACE_sRGB; - insert_node_between(input, conversion, node); + replace_sender(input, conversion); + connect_nodes(input, conversion); } // Re-sort topologically, and propagate the new information. @@ -1029,7 +1081,8 @@ void EffectChain::fix_internal_alpha(unsigned step) conversion = add_node(new AlphaDivisionEffect()); } conversion->output_alpha_type = desired_type; - insert_node_between(input, conversion, node); + replace_sender(input, conversion); + connect_nodes(input, conversion); } // Re-sort topologically, and propagate the new information. @@ -1080,14 +1133,14 @@ void EffectChain::fix_output_alpha() return; } if (output->output_alpha_type == ALPHA_PREMULTIPLIED && - output_alpha_format == OUTPUT_ALPHA_POSTMULTIPLIED) { + output_alpha_format == OUTPUT_ALPHA_FORMAT_POSTMULTIPLIED) { Node *conversion = add_node(new AlphaDivisionEffect()); connect_nodes(output, conversion); propagate_alpha(); propagate_gamma_and_color_space(); } if (output->output_alpha_type == ALPHA_POSTMULTIPLIED && - output_alpha_format == OUTPUT_ALPHA_PREMULTIPLIED) { + output_alpha_format == OUTPUT_ALPHA_FORMAT_PREMULTIPLIED) { Node *conversion = add_node(new AlphaMultiplicationEffect()); connect_nodes(output, conversion); propagate_alpha(); @@ -1153,7 +1206,7 @@ void EffectChain::fix_internal_gamma_by_asking_inputs(unsigned step) } // See if all inputs can give us linear gamma. If not, leave it. - std::vector nonlinear_inputs; + vector nonlinear_inputs; find_all_nonlinear_inputs(node, &nonlinear_inputs); assert(!nonlinear_inputs.empty()); @@ -1212,7 +1265,7 @@ void EffectChain::fix_internal_gamma_by_inserting_nodes(unsigned step) } // If not, go through each input that is not linear gamma, - // and insert a gamma conversion before it. + // and insert a gamma conversion after it. for (unsigned j = 0; j < node->incoming_links.size(); ++j) { Node *input = node->incoming_links[j]; assert(input->output_gamma_curve != GAMMA_INVALID); @@ -1222,7 +1275,8 @@ void EffectChain::fix_internal_gamma_by_inserting_nodes(unsigned step) Node *conversion = add_node(new GammaExpansionEffect()); CHECK(conversion->effect->set_int("source_curve", input->output_gamma_curve)); conversion->output_gamma_curve = GAMMA_LINEAR; - insert_node_between(input, conversion, node); + replace_sender(input, conversion); + connect_nodes(input, conversion); } // Re-sort topologically, and propagate the new information. @@ -1283,7 +1337,7 @@ void EffectChain::add_dither_if_needed() // multiple outputs right now). Node *EffectChain::find_output_node() { - std::vector output_nodes; + vector output_nodes; for (unsigned i = 0; i < nodes.size(); ++i) { Node *node = nodes[i]; if (node->disabled) { @@ -1299,6 +1353,10 @@ Node *EffectChain::find_output_node() void EffectChain::finalize() { + // Save the current locale, and set it to C, so that we can output decimal + // numbers with printf and be sure to get them in the format mandated by GLSL. + char *saved_locale = setlocale(LC_NUMERIC, "C"); + // Output the graph as it is before we do any conversions on it. output_dot("step0-start.dot"); @@ -1354,36 +1412,6 @@ void EffectChain::finalize() output_dot("step19-split-to-phases.dot"); - // If we have more than one phase, we need intermediate render-to-texture. - // Construct an FBO, and then as many textures as we need. - // We choose the simplest option of having one texture per output, - // since otherwise this turns into an (albeit simple) - // register allocation problem. - if (phases.size() > 1) { - glGenFramebuffers(1, &fbo); - - for (unsigned i = 0; i < phases.size() - 1; ++i) { - inform_input_sizes(phases[i]); - find_output_size(phases[i]); - - Node *output_node = phases[i]->effects.back(); - glGenTextures(1, &output_node->output_texture); - check_error(); - glBindTexture(GL_TEXTURE_2D, output_node->output_texture); - check_error(); - glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE); - check_error(); - glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE); - check_error(); - glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA16F_ARB, phases[i]->output_width, phases[i]->output_height, 0, GL_RGBA, GL_UNSIGNED_BYTE, NULL); - check_error(); - - output_node->output_texture_width = phases[i]->output_width; - output_node->output_texture_height = phases[i]->output_height; - } - inform_input_sizes(phases.back()); - } - for (unsigned i = 0; i < inputs.size(); ++i) { inputs[i]->finalize(); } @@ -1391,6 +1419,7 @@ void EffectChain::finalize() assert(phases[0]->inputs.empty()); finalized = true; + setlocale(LC_NUMERIC, saved_locale); } void EffectChain::render_to_fbo(GLuint dest_fbo, unsigned width, unsigned height) @@ -1399,6 +1428,7 @@ void EffectChain::render_to_fbo(GLuint dest_fbo, unsigned width, unsigned height // Save original viewport. GLuint x = 0, y = 0; + GLuint fbo = 0; if (width == 0 && height == 0) { GLint viewport[4]; @@ -1425,35 +1455,26 @@ void EffectChain::render_to_fbo(GLuint dest_fbo, unsigned width, unsigned height glLoadIdentity(); if (phases.size() > 1) { + glGenFramebuffers(1, &fbo); + check_error(); glBindFramebuffer(GL_FRAMEBUFFER, fbo); check_error(); } - std::set generated_mipmaps; + set generated_mipmaps; + + // We choose the simplest option of having one texture per output, + // since otherwise this turns into an (albeit simple) register allocation problem. + map output_textures; for (unsigned phase = 0; phase < phases.size(); ++phase) { - // See if the requested output size has changed. If so, we need to recreate - // the texture (and before we start setting up inputs). + // Find a texture for this phase. inform_input_sizes(phases[phase]); if (phase != phases.size() - 1) { find_output_size(phases[phase]); - Node *output_node = phases[phase]->effects.back(); - - if (output_node->output_texture_width != phases[phase]->output_width || - output_node->output_texture_height != phases[phase]->output_height) { - glActiveTexture(GL_TEXTURE0); - check_error(); - glBindTexture(GL_TEXTURE_2D, output_node->output_texture); - check_error(); - glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA16F_ARB, phases[phase]->output_width, phases[phase]->output_height, 0, GL_RGBA, GL_UNSIGNED_BYTE, NULL); - check_error(); - glBindTexture(GL_TEXTURE_2D, 0); - check_error(); - - output_node->output_texture_width = phases[phase]->output_width; - output_node->output_texture_height = phases[phase]->output_height; - } + GLuint tex_num = resource_pool->create_2d_texture(GL_RGBA16F_ARB, phases[phase]->output_width, phases[phase]->output_height); + output_textures.insert(make_pair(phases[phase], tex_num)); } glUseProgram(phases[phase]->glsl_program_num); @@ -1463,7 +1484,7 @@ void EffectChain::render_to_fbo(GLuint dest_fbo, unsigned width, unsigned height for (unsigned sampler = 0; sampler < phases[phase]->inputs.size(); ++sampler) { glActiveTexture(GL_TEXTURE0 + sampler); Node *input = phases[phase]->inputs[sampler]; - glBindTexture(GL_TEXTURE_2D, input->output_texture); + glBindTexture(GL_TEXTURE_2D, output_textures[input->phase]); check_error(); if (phases[phase]->input_needs_mipmaps) { if (generated_mipmaps.count(input) == 0) { @@ -1477,8 +1498,12 @@ void EffectChain::render_to_fbo(GLuint dest_fbo, unsigned width, unsigned height glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR); check_error(); } + glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE); + check_error(); + glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE); + check_error(); - std::string texture_name = std::string("tex_") + input->effect_id; + string texture_name = string("tex_") + phases[phase]->effect_ids[input]; glUniform1i(glGetUniformLocation(phases[phase]->glsl_program_num, texture_name.c_str()), sampler); check_error(); } @@ -1488,20 +1513,23 @@ void EffectChain::render_to_fbo(GLuint dest_fbo, unsigned width, unsigned height // Last phase goes to the output the user specified. glBindFramebuffer(GL_FRAMEBUFFER, dest_fbo); check_error(); + GLenum status = glCheckFramebufferStatusEXT(GL_FRAMEBUFFER_EXT); + assert(status == GL_FRAMEBUFFER_COMPLETE); glViewport(x, y, width, height); if (dither_effect != NULL) { CHECK(dither_effect->set_int("output_width", width)); CHECK(dither_effect->set_int("output_height", height)); } } else { - Node *output_node = phases[phase]->effects.back(); glFramebufferTexture2D( GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, - output_node->output_texture, + output_textures[phases[phase]], 0); check_error(); + GLenum status = glCheckFramebufferStatusEXT(GL_FRAMEBUFFER_EXT); + assert(status == GL_FRAMEBUFFER_COMPLETE); glViewport(0, 0, phases[phase]->output_width, phases[phase]->output_height); } @@ -1509,7 +1537,7 @@ void EffectChain::render_to_fbo(GLuint dest_fbo, unsigned width, unsigned height unsigned sampler_num = phases[phase]->inputs.size(); for (unsigned i = 0; i < phases[phase]->effects.size(); ++i) { Node *node = phases[phase]->effects[i]; - node->effect->set_gl_state(phases[phase]->glsl_program_num, node->effect_id, &sampler_num); + node->effect->set_gl_state(phases[phase]->glsl_program_num, phases[phase]->effect_ids[node], &sampler_num); check_error(); } @@ -1536,4 +1564,18 @@ void EffectChain::render_to_fbo(GLuint dest_fbo, unsigned width, unsigned height node->effect->clear_gl_state(); } } + + for (map::const_iterator texture_it = output_textures.begin(); + texture_it != output_textures.end(); + ++texture_it) { + resource_pool->release_2d_texture(texture_it->second); + } + + glBindFramebuffer(GL_FRAMEBUFFER, 0); + check_error(); + + if (fbo != 0) { + glDeleteFramebuffers(1, &fbo); + check_error(); + } }