GraphicsEngine

class GraphicsEngine

Bases: ReferenceCount

This class is the main interface to controlling the render process. There is typically only one GraphicsEngine in an application, and it synchronizes rendering to all all of the active windows; although it is possible to have multiple GraphicsEngine objects if multiple synchronicity groups are required.

The GraphicsEngine is responsible for managing the various cull and draw threads. The application simply calls engine->render_frame() and considers it done.

Inheritance diagram

Inheritance diagram of GraphicsEngine

explicit GraphicsEngine(Pipeline *pipeline = nullptr)

Creates a new GraphicsEngine object. The Pipeline is normally left to default to NULL, which indicates the global render pipeline, but it may be any Pipeline you choose.

bool add_window(GraphicsOutput *window, int sort)

This can be used to add a newly-created GraphicsOutput object (and its GSG) to the engine’s list of windows, and requests that it be opened. This shouldn’t be called by user code as make_output normally does this under the hood; it may be useful in esoteric cases in which a custom window object is used.

This can be called during the rendering loop, unlike make_output(); the window will be opened before the next frame begins rendering. Because it doesn’t call open_windows(), however, it’s not guaranteed that the window will succeed opening even if it returns true.

void dispatch_compute(LVecBase3i const &work_groups, ShaderAttrib const *sattr, GraphicsStateGuardian *gsg)

Asks the indicated GraphicsStateGuardian to dispatch the compute shader in the given ShaderAttrib using the given work group counts. This can act as an interface for running a one-off compute shader, without having to store it in the scene graph using a ComputeNode.

Since this requires a round-trip to the draw thread, it may require waiting for the current thread to finish rendering if it is called in a multithreaded environment. However, you can call this several consecutive times on different textures for little additional cost.

The return value is true if the operation is successful, false otherwise.

bool extract_texture_data(Texture *tex, GraphicsStateGuardian *gsg)

Asks the indicated GraphicsStateGuardian to retrieve the texture memory image of the indicated texture and store it in the texture’s ram_image field. The image can then be written to disk via Texture::write(), or otherwise manipulated on the CPU.

This is useful for retrieving the contents of a texture that has been somehow generated on the graphics card, instead of having been loaded the normal way via Texture::read() or Texture::load(). It is particularly useful for getting the data associated with a compressed texture image.

Since this requires a round-trip to the draw thread, it may require waiting for the current thread to finish rendering if it is called in a multithreaded environment. However, you can call this several consecutive times on different textures for little additional cost.

If the texture has not yet been loaded to the GSG in question, it will be loaded immediately.

The return value is true if the operation is successful, false otherwise.

void flip_frame(void)

Waits for all the threads that started drawing their last frame to finish drawing, and then flips all the windows. It is not usually necessary to call this explicitly, unless you need to see the previous frame right away.

bool get_auto_flip(void) const

Returns the current setting for the auto-flip flag. See set_auto_flip.

Loader *get_default_loader(void) const

Returns the Loader object that will be assigned to every GSG created with this GraphicsEngine. See GraphicsStateGuardian::set_loader().

static GraphicsEngine *get_global_ptr(void)
int get_num_windows(void) const

Returns the number of windows (or buffers) managed by the engine.

bool get_portal_cull(void) const

Returns the current setting for the portal culling flag.

ReMutex const &get_render_lock(void) const

Returns a ReMutex object that is held by the GraphicsEngine during the entire call to render_frame(). While you hold this lock you can be confident that no part of the frame will be rendered (at least by the app thread).

GraphicsThreadingModel get_threading_model(void) const

Returns the threading model that will be applied to future objects. See set_threading_model().

GraphicsOutput *get_window(int n) const

Returns the nth window or buffers managed by the engine, in sorted order.

bool is_empty(void) const

Returns true if there are no windows or buffers managed by the engine, false if there is at least one.

GraphicsOutput *make_buffer(GraphicsOutput *host, std::string const &name, int sort, int x_size, int y_size)
GraphicsOutput *make_buffer(GraphicsStateGuardian *gsg, std::string const &name, int sort, int x_size, int y_size)

Syntactic shorthand versions of make_output

Syntactic shorthand for make_output. This is the preferred way to create an offscreen buffer, when you already have an onscreen window or another buffer to start with. For the first parameter, pass an existing GraphicsOutput object, e.g. the main window; this allows the buffer to adapt itself to that window’s framebuffer properties, and allows maximum sharing of resources.

Syntactic shorthand for make_output. This flavor accepts a GSG rather than a GraphicsOutput as the first parameter, which is too limiting and disallows the possibility of creating a ParasiteBuffer if the user’s graphics hardware prefers that. It also attempts to request specific framebuffer properties and may therefore do a poorer job of sharing the GSG between the old buffer and the new.

For these reasons, this variant is a poor choice unless you are creating an offscreen buffer for the first time, without an onscreen window already in existence. If you already have an onscreen window, you should use the other flavor of make_buffer() instead, which accepts a GraphicsOutput as the first parameter.

GraphicsOutput *make_output(GraphicsPipe *pipe, std::string const &name, int sort, FrameBufferProperties const &fb_prop, WindowProperties const &win_prop, int flags, GraphicsStateGuardian *gsg = nullptr, GraphicsOutput *host = nullptr)
GraphicsOutput *make_parasite(GraphicsOutput *host, std::string const &name, int sort, int x_size, int y_size)

Syntactic shorthand for make_buffer.

void open_windows(void)

Fully opens (or closes) any windows that have recently been requested open or closed, without rendering any frames. It is not necessary to call this explicitly, since windows will be automatically opened or closed when the next frame is rendered, but you may call this if you want your windows now without seeing a frame go by.

void ready_flip(void)

Waits for all the threads that started drawing their last frame to finish drawing. Returns when all threads have actually finished drawing, as opposed to ‘sync_frame’ we seems to return once all draw calls have been submitted. Calling ‘flip_frame’ after this function should immediately cause a buffer flip. This function will only work in opengl right now, for all other graphics pipelines it will simply return immediately. In opengl it’s a bit of a hack: it will attempt to read a single pixel from the frame buffer to force the graphics card to finish drawing before it returns

void remove_all_windows(void)

Removes and closes all windows from the engine. This also cleans up and terminates any threads that have been started to service those windows.

bool remove_window(GraphicsOutput *window)

Removes the indicated window or offscreen buffer from the set of windows that will be processed when render_frame() is called. This also closes the window if it is open, and removes the window from its GraphicsPipe, allowing the window to be destructed if there are no other references to it. (However, the window may not be actually closed until next frame, if it is controlled by a sub-thread.)

The return value is true if the window was removed, false if it was not found.

Unlike remove_all_windows(), this function does not terminate any of the threads that may have been started to service this window; they are left running (since you might open a new window later on these threads). If your intention is to clean up before shutting down, it is better to call remove_all_windows() then to call remove_window() one at a time.

void render_frame(void)

Renders the next frame in all the registered windows, and flips all of the frame buffers.

void reset_all_windows(bool swapchain)

Resets the framebuffer of the current window. This is currently used by DirectX 8 only. It calls a reset_window function on each active window to release/create old/new framebuffer

void set_auto_flip(bool auto_flip)

Set this flag true to indicate the GraphicsEngine should automatically cause windows to sync and flip as soon as they have finished drawing, rather than waiting for all of the windows to finish drawing first so they can flip together.

This only affects the timing of when the flip occurs. If this is true (the default), the flip occurs before render_frame() returns. If this is false, the flip occurs whenever flip_frame() is called, or at the beginning of the next call to render_frame(), if flip_frame() is never called.

void set_default_loader(Loader *loader)

Sets the Loader object that will be assigned to every GSG created with this GraphicsEngine. See GraphicsStateGuardian::set_loader().

void set_portal_cull(bool value)

Set this flag true to indicate the GraphicsEngine should start portal culling

void set_threading_model(GraphicsThreadingModel const &threading_model)

Specifies how future objects created via make_gsg(), make_buffer(), and make_output() will be threaded. This does not affect any already-created objects.

void sync_frame(void)

Waits for all the threads that started drawing their last frame to finish drawing. The windows are not yet flipped when this returns; see also flip_frame(). It is not usually necessary to call this explicitly, unless you need to see the previous frame right away.