#define GL_GLEXT_PROTOTYPES 1
-#include <stdio.h>
+#include <GL/glew.h>
+#include <assert.h>
+#include <locale.h>
#include <math.h>
+#include <stddef.h>
+#include <stdio.h>
+#include <stdlib.h>
#include <string.h>
-#include <assert.h>
-#include <GL/glew.h>
-
#include <algorithm>
#include <set>
#include <stack>
#include <vector>
-#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)
+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()
{
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;
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;
nodes.push_back(node);
node_map[effect] = node;
+ effect->inform_added(this);
return node;
}
Effect *EffectChain::add_effect(Effect *effect, const std::vector<Effect *> &inputs)
{
+ assert(!finalized);
assert(inputs.size() == effect->num_inputs());
Node *node = add_node(effect);
for (unsigned i = 0; i < inputs.size(); ++i) {
const std::vector<Node *> &inputs,
const std::vector<Node *> &effects)
{
+ Phase *phase = new Phase;
assert(!effects.empty());
// Deduplicate the inputs.
// 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(std::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 += std::string("uniform sampler2D tex_") + effect_id + ";\n";
+ frag_shader += std::string("vec4 ") + effect_id + "(vec2 tc) {\n";
+ frag_shader += "\treturn texture2D(tex_" + std::string(effect_id) + ", tc);\n";
frag_shader += "}\n";
frag_shader += "\n";
}
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(std::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 += std::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 += std::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) {
CHECK(node->effect->set_int("needs_mipmaps", input_needs_mipmaps));
}
}
- frag_shader += std::string("#define INPUT ") + sorted_effects.back()->effect_id + "\n";
+ frag_shader += std::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;
// 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<Node *> completed_effects;
// 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);
}
}
-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;
}
}
}
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);
}
// 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) {
}
Input *input = static_cast<Input *>(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;
}
}
- assert(best_width != 0);
- phase->output_width = best_width;
- phase->output_height = best_width * aspect_denom / aspect_nom;
+
+ 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<Input *>(effect);
+ size_rectangle_to_fit(input->get_width(), input->get_height(), &output_width, &output_height);
+ }
+ 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()
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);
}
// 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
// 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
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 {
}
// 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);
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.
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.
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();
}
// 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);
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.
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");
// 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);
+ output_node->output_texture = resource_pool->create_2d_texture(GL_RGBA16F_ARB, phases[i]->output_width, phases[i]->output_height);
check_error();
glBindTexture(GL_TEXTURE_2D, output_node->output_texture);
check_error();
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;
assert(phases[0]->inputs.empty());
finalized = true;
+ setlocale(LC_NUMERIC, saved_locale);
}
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];
glLoadIdentity();
if (phases.size() > 1) {
+ glGenFramebuffers(1, &fbo);
+ check_error();
glBindFramebuffer(GL_FRAMEBUFFER, fbo);
check_error();
}
check_error();
}
- std::string texture_name = std::string("tex_") + input->effect_id;
+ std::string texture_name = std::string("tex_") + phases[phase]->effect_ids[input];
glUniform1i(glGetUniformLocation(phases[phase]->glsl_program_num, texture_name.c_str()), sampler);
check_error();
}
// 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));
output_node->output_texture,
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);
}
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();
}
node->effect->clear_gl_state();
}
}
+
+ glBindFramebuffer(GL_FRAMEBUFFER, 0);
+ check_error();
+
+ if (fbo != 0) {
+ glDeleteFramebuffers(1, &fbo);
+ check_error();
+ }
}
projects / movit / blobdiff