namespace movit {
+namespace {
+
+// An effect that does nothing.
+class IdentityEffect : public Effect {
+public:
+ IdentityEffect() {}
+ virtual string effect_type_id() const { return "IdentityEffect"; }
+ string output_fragment_shader() { return read_file("identity.frag"); }
+};
+
+} // namespace
+
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),
+ num_output_color_ycbcr(0),
dither_effect(NULL),
+ ycbcr_conversion_effect_node(NULL),
+ intermediate_format(GL_RGBA16F),
+ intermediate_transformation(NO_FRAMEBUFFER_TRANSFORMATION),
num_dither_bits(0),
output_origin(OUTPUT_ORIGIN_BOTTOM_LEFT),
finalized(false),
}
void EffectChain::add_ycbcr_output(const ImageFormat &format, OutputAlphaFormat alpha_format,
- const YCbCrFormat &ycbcr_format, YCbCrOutputSplitting output_splitting)
+ const YCbCrFormat &ycbcr_format, YCbCrOutputSplitting output_splitting,
+ GLenum output_type)
{
assert(!finalized);
- assert(!output_color_ycbcr);
+ assert(num_output_color_ycbcr < 2);
output_format = format;
output_alpha_format = alpha_format;
- output_color_ycbcr = true;
- output_ycbcr_format = ycbcr_format;
- output_ycbcr_splitting = output_splitting;
+
+ if (num_output_color_ycbcr == 1) {
+ // Check that the format is the same.
+ assert(output_ycbcr_format.luma_coefficients == ycbcr_format.luma_coefficients);
+ assert(output_ycbcr_format.full_range == ycbcr_format.full_range);
+ assert(output_ycbcr_format.num_levels == ycbcr_format.num_levels);
+ assert(output_ycbcr_format.chroma_subsampling_x == 1);
+ assert(output_ycbcr_format.chroma_subsampling_y == 1);
+ assert(output_ycbcr_type == output_type);
+ } else {
+ output_ycbcr_format = ycbcr_format;
+ output_ycbcr_type = output_type;
+ }
+ output_ycbcr_splitting[num_output_color_ycbcr++] = output_splitting;
assert(ycbcr_format.chroma_subsampling_x == 1);
assert(ycbcr_format.chroma_subsampling_y == 1);
}
+void EffectChain::change_ycbcr_output_format(const YCbCrFormat &ycbcr_format)
+{
+ assert(num_output_color_ycbcr > 0);
+ assert(output_ycbcr_format.chroma_subsampling_x == 1);
+ assert(output_ycbcr_format.chroma_subsampling_y == 1);
+
+ output_ycbcr_format = ycbcr_format;
+ if (finalized) {
+ YCbCrConversionEffect *effect = (YCbCrConversionEffect *)(ycbcr_conversion_effect_node->effect);
+ effect->change_output_format(ycbcr_format);
+ }
+}
+
Node *EffectChain::add_node(Effect *effect)
{
for (unsigned i = 0; i < nodes.size(); ++i) {
void EffectChain::compile_glsl_program(Phase *phase)
{
- string frag_shader_header = read_version_dependent_file("header", "frag");
+ string frag_shader_header;
+ if (phase->is_compute_shader) {
+ frag_shader_header = read_file("header.compute");
+ } else {
+ frag_shader_header = read_version_dependent_file("header", "frag");
+ }
string frag_shader = "";
// Create functions and uniforms for all the texture inputs that we need.
frag_shader += string("uniform sampler2D tex_") + effect_id + ";\n";
frag_shader += string("vec4 ") + effect_id + "(vec2 tc) {\n";
- frag_shader += "\treturn tex2D(tex_" + string(effect_id) + ", tc);\n";
+ frag_shader += "\tvec4 tmp = tex2D(tex_" + string(effect_id) + ", tc);\n";
+
+ if (intermediate_transformation == SQUARE_ROOT_FRAMEBUFFER_TRANSFORMATION &&
+ phase->inputs[i]->output_node->output_gamma_curve == GAMMA_LINEAR) {
+ frag_shader += "\ttmp.rgb *= tmp.rgb;\n";
+ }
+
+ frag_shader += "\treturn tmp;\n";
frag_shader += "}\n";
frag_shader += "\n";
frag_shader += "\n";
frag_shader += string("#define FUNCNAME ") + effect_id + "\n";
+ if (node->effect->is_compute_shader()) {
+ frag_shader += string("#define NORMALIZE_TEXTURE_COORDS(tc) ((tc) * ") + effect_id + "_inv_output_size + " + effect_id + "_output_texcoord_adjust)\n";
+ }
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) {
frag_shader += "#undef INPUT\n";
// 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) {
+ if (phase->output_node->outgoing_links.empty() && num_output_color_ycbcr > 0) {
+ switch (output_ycbcr_splitting[0]) {
case YCBCR_OUTPUT_INTERLEAVED:
// No #defines set.
frag_shader_outputs.push_back("FragColor");
assert(false);
}
+ if (num_output_color_ycbcr > 1) {
+ switch (output_ycbcr_splitting[1]) {
+ case YCBCR_OUTPUT_INTERLEAVED:
+ frag_shader += "#define SECOND_YCBCR_OUTPUT_INTERLEAVED 1\n";
+ frag_shader_outputs.push_back("YCbCr2");
+ break;
+ case YCBCR_OUTPUT_SPLIT_Y_AND_CBCR:
+ frag_shader += "#define SECOND_YCBCR_OUTPUT_SPLIT_Y_AND_CBCR 1\n";
+ frag_shader_outputs.push_back("Y2");
+ frag_shader_outputs.push_back("Chroma2");
+ break;
+ case YCBCR_OUTPUT_PLANAR:
+ frag_shader += "#define SECOND_YCBCR_OUTPUT_PLANAR 1\n";
+ frag_shader_outputs.push_back("Y2");
+ frag_shader_outputs.push_back("Cb2");
+ frag_shader_outputs.push_back("Cr2");
+ 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
frag_shader_outputs.push_back("RGBA");
}
}
- frag_shader.append(read_file("footer.frag"));
+
+ // If we're bouncing to a temporary texture, signal transformation if desired.
+ if (!phase->output_node->outgoing_links.empty()) {
+ if (intermediate_transformation == SQUARE_ROOT_FRAMEBUFFER_TRANSFORMATION &&
+ phase->output_node->output_gamma_curve == GAMMA_LINEAR) {
+ frag_shader += "#define SQUARE_ROOT_TRANSFORMATION 1\n";
+ }
+ }
+
+ if (phase->is_compute_shader) {
+ frag_shader.append(read_file("footer.compute"));
+ phase->output_node->effect->register_uniform_vec2("inv_output_size", (float *)&phase->inv_output_size);
+ phase->output_node->effect->register_uniform_vec2("output_texcoord_adjust", (float *)&phase->output_texcoord_adjust);
+ } else {
+ 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
Node *node = phase->effects[i];
Effect *effect = node->effect;
const string effect_id = phase->effect_ids[node];
+ extract_uniform_declarations(effect->uniforms_image2d, "image2D", effect_id, &phase->uniforms_image2d, &frag_shader_uniforms);
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);
vert_shader[pos + needle.size() - 1] = '1';
}
- phase->glsl_program_num = resource_pool->compile_glsl_program(vert_shader, frag_shader, frag_shader_outputs);
+ if (phase->is_compute_shader) {
+ phase->glsl_program_num = resource_pool->compile_glsl_compute_program(frag_shader);
+
+ Uniform<int> uniform;
+ uniform.name = "outbuf";
+ uniform.value = &phase->outbuf_image_unit;
+ uniform.prefix = "tex";
+ uniform.num_values = 1;
+ uniform.location = -1;
+ phase->uniforms_image2d.push_back(uniform);
+ } else {
+ 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) {
}
// Collect the resulting location numbers for each uniform.
+ collect_uniform_locations(phase->glsl_program_num, &phase->uniforms_image2d);
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);
Phase *phase = new Phase;
phase->output_node = output;
+ phase->is_compute_shader = output->effect->is_compute_shader();
// If the output effect has one-to-one sampling, we try to trace this
// status down through the dependency chain. This is important in case
start_new_phase = true;
}
+ // Compute shaders currently always end phases.
+ // (We might loosen this up in some cases in the future.)
+ if (deps[i]->effect->is_compute_shader()) {
+ start_new_phase = true;
+ }
+
// Propagate information about needing mipmaps down the chain,
// breaking the phase if we notice an incompatibility.
//
if (alpha_handling == Effect::INPUT_AND_OUTPUT_PREMULTIPLIED_ALPHA ||
alpha_handling == Effect::INPUT_PREMULTIPLIED_ALPHA_KEEP_BLANK) {
+ // This combination (requiring premultiplied alpha, but _not_ requiring
+ // linear light) is illegal, since the combination of premultiplied alpha
+ // and nonlinear inputs is meaningless.
+ assert(node->effect->needs_linear_light());
+
// If the effect has asked for premultiplied alpha, check that it has got it.
if (any_postmultiplied) {
node->output_alpha_type = ALPHA_INVALID;
// gamma-encoded data.
void EffectChain::add_ycbcr_conversion_if_needed()
{
- assert(output_color_rgba || output_color_ycbcr);
- if (!output_color_ycbcr) {
+ assert(output_color_rgba || num_output_color_ycbcr > 0);
+ if (num_output_color_ycbcr == 0) {
return;
}
Node *output = find_output_node();
- Node *ycbcr = add_node(new YCbCrConversionEffect(output_ycbcr_format));
- connect_nodes(output, ycbcr);
+ ycbcr_conversion_effect_node = add_node(new YCbCrConversionEffect(output_ycbcr_format, output_ycbcr_type));
+ connect_nodes(output, ycbcr_conversion_effect_node);
}
// If the user has requested dither, add a DitherEffect right at the end
dither_effect = dither->effect;
}
+// Compute shaders can't output to the framebuffer, so if the last
+// phase ends in a compute shader, add a dummy phase at the end that
+// only blits directly from the temporary texture.
+//
+// TODO: Add an API for rendering directly to textures, for the cases
+// where we're only rendering to an FBO anyway.
+void EffectChain::add_dummy_effect_if_needed()
+{
+ Node *output = find_output_node();
+ if (output->effect->is_compute_shader()) {
+ Node *dummy = add_node(new IdentityEffect());
+ connect_nodes(output, dummy);
+ }
+}
+
// Find the output node. This is, simply, one that has no outgoing links.
// If there are multiple ones, the graph is malformed (we do not support
// multiple outputs right now).
output_dot("step18-before-dither.dot");
add_dither_if_needed();
- output_dot("step19-final.dot");
+ output_dot("step19-before-dummy-effect.dot");
+ add_dummy_effect_if_needed();
+
+ output_dot("step20-final.dot");
// Construct all needed GLSL programs, starting at the output.
// We need to keep track of which effects have already been computed,
map<Node *, Phase *> completed_effects;
construct_phase(find_output_node(), &completed_effects);
- output_dot("step20-split-to-phases.dot");
+ output_dot("step21-split-to-phases.dot");
assert(phases[0]->inputs.empty());
glDisable(GL_DITHER);
check_error();
+ const bool final_srgb = glIsEnabled(GL_FRAMEBUFFER_SRGB);
+ check_error();
+ bool current_srgb = final_srgb;
+
// Save original viewport.
GLuint x = 0, y = 0;
glDepthMask(GL_FALSE);
check_error();
- // 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();
- glBindBuffer(GL_ARRAY_BUFFER, vbo);
- check_error();
- set<GLint> bound_attribute_indices;
-
set<Phase *> generated_mipmaps;
// We choose the simplest option of having one texture per output,
CHECK(dither_effect->set_int("output_height", height));
}
}
- execute_phase(phase, phase_num == phases.size() - 1, &bound_attribute_indices, &output_textures, &generated_mipmaps);
+ bool last_phase = (phase_num == phases.size() - 1);
+
+ // Enable sRGB rendering for intermediates in case we are
+ // rendering to an sRGB format.
+ bool needs_srgb = last_phase ? final_srgb : true;
+ if (needs_srgb && !current_srgb) {
+ glEnable(GL_FRAMEBUFFER_SRGB);
+ check_error();
+ current_srgb = true;
+ } else if (!needs_srgb && current_srgb) {
+ glDisable(GL_FRAMEBUFFER_SRGB);
+ check_error();
+ current_srgb = true;
+ }
+
+ execute_phase(phase, last_phase, &output_textures, &generated_mipmaps);
if (do_phase_timing) {
glEndQuery(GL_TIME_ELAPSED);
}
check_error();
glBindVertexArray(0);
check_error();
- glDeleteVertexArrays(1, &vao);
- check_error();
if (do_phase_timing) {
// Get back the timer queries.
}
void EffectChain::execute_phase(Phase *phase, bool last_phase,
- set<GLint> *bound_attribute_indices,
map<Phase *, GLuint> *output_textures,
set<Phase *> *generated_mipmaps)
{
if (!last_phase) {
find_output_size(phase);
- GLuint tex_num = resource_pool->create_2d_texture(GL_RGBA16F, phase->output_width, phase->output_height);
+ GLuint tex_num = resource_pool->create_2d_texture(intermediate_format, phase->output_width, phase->output_height);
output_textures->insert(make_pair(phase, tex_num));
}
- glUseProgram(phase->glsl_program_num);
- check_error();
-
// Set up RTT inputs for this phase.
for (unsigned sampler = 0; sampler < phase->inputs.size(); ++sampler) {
glActiveTexture(GL_TEXTURE0 + sampler);
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.)
- if (!last_phase) {
- fbo = resource_pool->create_fbo((*output_textures)[phase]);
- glBindFramebuffer(GL_FRAMEBUFFER, fbo);
- glViewport(0, 0, phase->output_width, phase->output_height);
+ GLuint instance_program_num = resource_pool->use_glsl_program(phase->glsl_program_num);
+ check_error();
+
+ // And now the output.
+ if (phase->is_compute_shader) {
+ // This is currently the only place where we use image units,
+ // so we can always use 0.
+ phase->outbuf_image_unit = 0;
+ glBindImageTexture(phase->outbuf_image_unit, (*output_textures)[phase], 0, GL_FALSE, 0, GL_WRITE_ONLY, GL_RGBA16F);
+ check_error();
+ phase->inv_output_size.x = 1.0f / phase->output_width;
+ phase->inv_output_size.y = 1.0f / phase->output_height;
+ phase->output_texcoord_adjust.x = 0.5f / phase->output_width;
+ phase->output_texcoord_adjust.y = 0.5f / phase->output_height;
+ } else {
+ // (Already set up for us if it is the last phase.)
+ if (!last_phase) {
+ fbo = resource_pool->create_fbo((*output_textures)[phase]);
+ glBindFramebuffer(GL_FRAMEBUFFER, fbo);
+ glViewport(0, 0, phase->output_width, phase->output_height);
+ }
}
// Give the required parameters to all the effects.
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(phase->glsl_program_num, phase->effect_ids[node], &sampler_num);
+ node->effect->set_gl_state(instance_program_num, phase->effect_ids[node], &sampler_num);
check_error();
if (node->effect->is_single_texture()) {
}
}
- // Uniforms need to come after set_gl_state(), since they can be updated
- // from there.
- setup_uniforms(phase);
- // 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;
- }
- }
+ if (phase->is_compute_shader) {
+ unsigned x, y, z;
+ phase->output_node->effect->get_compute_dimensions(phase->output_width, phase->output_height, &x, &y, &z);
- // 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);
- }
- }
+ // Uniforms need to come after set_gl_state() _and_ get_compute_dimensions(),
+ // since they can be updated from there.
+ setup_uniforms(phase);
+ glDispatchCompute(x, y, z);
+ } else {
+ // Uniforms need to come after set_gl_state(), since they can be updated
+ // from there.
+ setup_uniforms(phase);
- glDrawArrays(GL_TRIANGLES, 0, 3);
- check_error();
+ // Bind the vertex data.
+ GLuint vao = resource_pool->create_vec2_vao(phase->attribute_indexes, vbo);
+ glBindVertexArray(vao);
+
+ glDrawArrays(GL_TRIANGLES, 0, 3);
+ check_error();
+
+ resource_pool->release_vec2_vao(vao);
+ }
for (unsigned i = 0; i < phase->effects.size(); ++i) {
Node *node = phase->effects[i];
node->effect->clear_gl_state();
}
- if (!last_phase) {
+ resource_pool->unuse_glsl_program(instance_program_num);
+
+ if (!last_phase && !phase->is_compute_shader) {
resource_pool->release_fbo(fbo);
}
}
void EffectChain::setup_uniforms(Phase *phase)
{
// TODO: Use UBO blocks.
+ for (size_t i = 0; i < phase->uniforms_image2d.size(); ++i) {
+ const Uniform<int> &uniform = phase->uniforms_image2d[i];
+ if (uniform.location != -1) {
+ glUniform1iv(uniform.location, uniform.num_values, uniform.value);
+ }
+ }
for (size_t i = 0; i < phase->uniforms_sampler2d.size(); ++i) {
const Uniform<int> &uniform = phase->uniforms_sampler2d[i];
if (uniform.location != -1) {
projects / movit / blobdiff