X-Git-Url: https://git.sesse.net/?p=movit;a=blobdiff_plain;f=effect_chain.h;h=716a970a3ab59748e99d94da43b2a9239fdca871;hp=ffa938947a6de8311ca8f17681b70ac4e9f4ca24;hb=094bfbb35f1af70ef07890ac7735006cdc195cca;hpb=aa1fce6c5356faa368758b15b3ad257c0f7bda0b diff --git a/effect_chain.h b/effect_chain.h index ffa9389..716a970 100644 --- a/effect_chain.h +++ b/effect_chain.h @@ -178,19 +178,32 @@ struct Phase { std::vector effects; // In order. unsigned output_width, output_height, virtual_output_width, virtual_output_height; + // Whether this phase is compiled as a compute shader, ie., the last effect is + // marked as one. + bool is_compute_shader; + + // If , which image unit the output buffer is bound to. + // This is used as source for a Uniform below. + int outbuf_image_unit; + + // These are used in transforming from unnormalized to normalized coordinates + // in compute shaders. + Point2D inv_output_size, output_texcoord_adjust; + // Identifier used to create unique variables in GLSL. // Unique per-phase to increase cacheability of compiled shaders. std::map effect_ids; // Uniforms for this phase; combined from all the effects. - std::vector > uniforms_sampler2d; - std::vector > uniforms_bool; - std::vector > uniforms_int; - std::vector > uniforms_float; - std::vector > uniforms_vec2; - std::vector > uniforms_vec3; - std::vector > uniforms_vec4; - std::vector > uniforms_mat3; + std::vector> uniforms_image2d; + std::vector> uniforms_sampler2d; + std::vector> uniforms_bool; + std::vector> uniforms_int; + std::vector> uniforms_float; + std::vector> uniforms_vec2; + std::vector> uniforms_vec3; + std::vector> uniforms_vec4; + std::vector> uniforms_mat3; // For measurement of GPU time used. std::list timer_query_objects_running; @@ -203,11 +216,11 @@ class EffectChain { public: // Aspect: e.g. 16.0f, 9.0f for 16:9. // resource_pool is a pointer to a ResourcePool with which to share shaders - // and other resources (see resource_pool.h). If NULL (the default), + // and other resources (see resource_pool.h). If nullptr (the default), // will create its own that is not shared with anything else. Does not take // ownership of the passed-in ResourcePool, but will naturally take ownership // of its own internal one if created. - EffectChain(float aspect_nom, float aspect_denom, ResourcePool *resource_pool = NULL); + EffectChain(float aspect_nom, float aspect_denom, ResourcePool *resource_pool = nullptr); ~EffectChain(); // User API: @@ -264,7 +277,7 @@ public: void add_output(const ImageFormat &format, OutputAlphaFormat alpha_format); // Adds an YCbCr output. Note that you can only have at most two Y'CbCr - // outputs, and they must have the same . + // outputs, and they must have the same and . // (This limitation may be lifted in the future, to allow e.g. simultaneous // 8- and 10-bit output. Currently, multiple Y'CbCr outputs are only // useful in some very limited circumstances, like if one texture goes @@ -272,13 +285,19 @@ public: // // Only 4:4:4 output is supported due to fragment shader limitations, // so chroma_subsampling_x and chroma_subsampling_y must both be 1. + // should match the data type of the FBO you are rendering to, + // so that if you use 16-bit output (GL_UNSIGNED_SHORT), you will get + // 8-, 10- or 12-bit output correctly as determined by . + // Using e.g. ycbcr_format.num_levels == 1024 with GL_UNSIGNED_BYTE is + // nonsensical and invokes undefined behavior. // // If you have both RGBA and Y'CbCr output(s), the RGBA output will come // in the last draw buffer. Also, and must be // identical between the two. void add_ycbcr_output(const ImageFormat &format, OutputAlphaFormat alpha_format, const YCbCrFormat &ycbcr_format, - YCbCrOutputSplitting output_splitting = YCBCR_OUTPUT_INTERLEAVED); + YCbCrOutputSplitting output_splitting = YCBCR_OUTPUT_INTERLEAVED, + GLenum output_type = GL_UNSIGNED_BYTE); // Change Y'CbCr output format. (This can be done also after finalize()). // Note that you are not allowed to change subsampling parameters; @@ -288,6 +307,11 @@ public: // Set number of output bits, to scale the dither. // 8 is the right value for most outputs. + // + // Special note for 10- and 12-bit Y'CbCr packed into GL_UNSIGNED_SHORT: + // This is relative to the actual output, not the logical one, so you should + // specify 16 here, not 10 or 12. + // // The default, 0, is a special value that means no dither. void set_dither_bits(unsigned num_bits) { @@ -353,6 +377,9 @@ public: void reset_phase_timing(); void print_phase_timing(); + // Note: If you already know the width and height of the viewport, + // calling render_to_fbo() directly will be slightly more efficient, + // as it saves it from getting it from OpenGL. void render_to_screen() { render_to_fbo(0, 0, 0); @@ -362,9 +389,30 @@ public: // the current viewport. void render_to_fbo(GLuint fbo, unsigned width, unsigned height); + // Render the effect chain to the given set of textures. This is equivalent + // to render_to_fbo() with a freshly created FBO bound to the given textures, + // except that it is more efficient if the last phase contains a compute shader. + // Thus, prefer this to render_to_fbo() where possible. + // + // Only one destination texture is supported. This restriction will be lifted + // in the future. + // + // All destination textures must be exactly of size x , + // and must either come from the same ResourcePool the effect uses, or outlive + // the EffectChain (otherwise, we could be allocating FBOs that end up being + // stale). Textures must also have valid state; in particular, they must either + // be mipmap complete or have a non-mipmapped minification mode. + // + // width and height can not be zero. + struct DestinationTexture { + GLuint texnum; + GLenum format; + }; + void render_to_texture(const std::vector &destinations, unsigned width, unsigned height); + Effect *last_added_effect() { if (nodes.empty()) { - return NULL; + return nullptr; } else { return nodes.back()->effect; } @@ -430,10 +478,19 @@ private: // as the last effect. Also pushes all phases in order onto . Phase *construct_phase(Node *output, std::map *completed_effects); + // Do the actual rendering of the chain. If is not (GLuint)-1, + // renders to that FBO. If is non-empty, render to that set + // of textures (last phase, save for the dummy phase, must be a compute shader), + // with x/y ignored. Having both set is an error. + void render(GLuint dest_fbo, const std::vector &destinations, + unsigned x, unsigned y, unsigned width, unsigned height); + // Execute one phase, ie. set up all inputs, effects and outputs, and render the quad. - void execute_phase(Phase *phase, bool last_phase, - std::set *bound__attribute_indices, - std::map *output_textures, + // If is empty, uses whatever output is current (and the phase must not be + // a compute shader). + void execute_phase(Phase *phase, + const std::map &output_textures, + const std::vector &destinations, std::set *generated_mipmaps); // Set up uniforms for one phase. The program must already be bound. @@ -486,6 +543,7 @@ private: void fix_output_gamma(); void add_ycbcr_conversion_if_needed(); void add_dither_if_needed(); + void add_dummy_effect_if_needed(); float aspect_nom, aspect_denom; ImageFormat output_format; @@ -494,6 +552,7 @@ private: bool output_color_rgba; int num_output_color_ycbcr; // Max 2. YCbCrFormat output_ycbcr_format; // If num_output_color_ycbcr is > 0. + GLenum output_ycbcr_type; // If num_output_color_ycbcr is > 0. YCbCrOutputSplitting output_ycbcr_splitting[2]; // If num_output_color_ycbcr is > N. std::vector nodes; @@ -511,6 +570,12 @@ private: bool finalized; GLuint vbo; // Contains vertex and texture coordinate data. + // Whether the last effect (which will then be in a phase all by itself) + // is a dummy effect that is only added because the last phase uses a compute + // shader, which cannot output directly to the backbuffer. This means that + // the phase can be skipped if we are _not_ rendering to the backbuffer. + bool has_dummy_effect = false; + ResourcePool *resource_pool; bool owns_resource_pool;