Files
MapLibreParticleLab/include/mbgl/gfx/renderer_backend.hpp

73 lines
1.9 KiB
C++

#pragma once
#include <mbgl/util/util.hpp>
#include <memory>
#include <mutex>
namespace mbgl {
namespace gfx {
class Context;
class Renderable;
class BackendScope;
// We can make some optimizations if we know that the drawing context is not shared with other code.
enum class ContextMode : bool {
Unique,
Shared,
};
class RendererBackend {
protected:
explicit RendererBackend(ContextMode);
public:
virtual ~RendererBackend();
RendererBackend(const RendererBackend&) = delete;
RendererBackend& operator=(const RendererBackend&) = delete;
// Returns the device's context.
Context& getContext();
template <typename T>
T& getContext() {
return static_cast<T&>(getContext());
}
bool contextIsShared() const {
return contextMode == ContextMode::Shared;
}
// Returns a reference to the default surface that should be rendered on.
virtual Renderable& getDefaultRenderable() = 0;
protected:
virtual std::unique_ptr<Context> createContext() = 0;
// Called when the backend's GL context needs to be made active or inactive. These are called,
// as a matched pair, exclusively through BackendScope, in two situations:
//
// 1. When releasing GL resources during Renderer destruction
// (Including calling CustomLayerHost::deinitialize during RenderCustomLayer destruction)
// 2. When renderering through Renderer::render()
// (Including calling CustomLayerHost::initialize for newly added custom layers and
// CustomLayerHost::deinitialize on layer removal)
virtual void activate() = 0;
virtual void deactivate() = 0;
protected:
std::unique_ptr<Context> context;
const ContextMode contextMode;
std::once_flag initialized;
friend class BackendScope;
};
MBGL_CONSTEXPR bool operator==(const RendererBackend& a, const RendererBackend& b) {
return &a == &b;
}
} // namespace gfx
} // namespace mbgl