/*
* Copyright 2014 Google Inc.
*
* Use of this source code is governed by a BSD-style license that can be
* found in the LICENSE file.
*/
#ifndef SkDrawable_DEFINED
#define SkDrawable_DEFINED
#include "SkFlattenable.h"
#include "SkImageInfo.h"
#include "SkScalar.h"
class GrBackendDrawableInfo;
class SkCanvas;
class SkMatrix;
class SkPicture;
enum class GrBackendApi : unsigned;
struct SkRect;
/**
* Base-class for objects that draw into SkCanvas.
*
* The object has a generation ID, which is guaranteed to be unique across all drawables. To
* allow for clients of the drawable that may want to cache the results, the drawable must
* change its generation ID whenever its internal state changes such that it will draw differently.
*/
class SK_API SkDrawable : public SkFlattenable {
public:
/**
* Draws into the specified content. The drawing sequence will be balanced upon return
* (i.e. the saveLevel() on the canvas will match what it was when draw() was called,
* and the current matrix and clip settings will not be changed.
*/
void draw(SkCanvas*, const SkMatrix* = nullptr);
void draw(SkCanvas*, SkScalar x, SkScalar y);
/**
* When using the GPU backend it is possible for a drawable to execute using the underlying 3D
* API rather than the SkCanvas API. It does so by creating a GpuDrawHandler. The GPU backend
* is deferred so the handler will be given access to the 3D API at the correct point in the
* drawing stream as the GPU backend flushes. Since the drawable may mutate, each time it is
* drawn to a GPU-backed canvas a new handler is snapped, representing the drawable's state at
* the time of the snap.
*
* When the GPU backend flushes to the 3D API it will call the draw method on the
* GpuDrawHandler. At this time the drawable may add commands to the stream of GPU commands for
* the unerlying 3D API. The draw function takes a GrBackendDrawableInfo which contains
* information about the current state of 3D API which the caller must respect. See
* GrBackendDrawableInfo for more specific details on what information is sent and the
* requirements for different 3D APIs.
*
* Additionaly there may be a slight delay from when the drawable adds its commands to when
* those commands are actually submitted to the GPU. Thus the drawable or GpuDrawHandler is
* required to keep any resources that are used by its added commands alive and valid until
* those commands are submitted to the GPU. The GpuDrawHandler will be kept alive and then
* deleted once the commands are submitted to the GPU. The dtor of the GpuDrawHandler is the
* signal to the drawable that the commands have all been submitted. Different 3D APIs may have
* additional requirements for certain resources which require waiting for the GPU to finish
* all work on those resources before reusing or deleting them. In this case, the drawable can
* use the dtor call of the GpuDrawHandler to add a fence to the GPU to track when the GPU work
* has completed.
*
* Currently this is only supported for the GPU Vulkan backend.
*/
class GpuDrawHandler {
public:
virtual ~GpuDrawHandler() {}
virtual void draw(const GrBackendDrawableInfo&) {}
};
/**
* Snaps off a GpuDrawHandler to represent the state of the SkDrawable at the time the snap is
* called. This is used for executing GPU backend specific draws intermixed with normal Skia GPU
* draws. The GPU API, which will be used for the draw, as well as the full matrix, device clip
* bounds and imageInfo of the target buffer are passed in as inputs.
*/
std::unique_ptr<GpuDrawHandler> snapGpuDrawHandler(GrBackendApi backendApi,
const SkMatrix& matrix,
const SkIRect& clipBounds,
const SkImageInfo& bufferInfo) {
return this->onSnapGpuDrawHandler(backendApi, matrix, clipBounds, bufferInfo);
}
SkPicture* newPictureSnapshot();
/**
* Return a unique value for this instance. If two calls to this return the same value,
* it is presumed that calling the draw() method will render the same thing as well.
*
* Subclasses that change their state should call notifyDrawingChanged() to ensure that
* a new value will be returned the next time it is called.
*/
uint32_t getGenerationID();
/**
* Return the (conservative) bounds of what the drawable will draw. If the drawable can
* change what it draws (e.g. animation or in response to some external change), then this
* must return a bounds that is always valid for all possible states.
*/
SkRect getBounds();
/**
* Calling this invalidates the previous generation ID, and causes a new one to be computed
* the next time getGenerationID() is called. Typically this is called by the object itself,
* in response to its internal state changing.
*/
void notifyDrawingChanged();
static SkFlattenable::Type GetFlattenableType() {
return kSkDrawable_Type;
}
SkFlattenable::Type getFlattenableType() const override {
return kSkDrawable_Type;
}
static sk_sp<SkDrawable> Deserialize(const void* data, size_t size,
const SkDeserialProcs* procs = nullptr) {
return sk_sp<SkDrawable>(static_cast<SkDrawable*>(
SkFlattenable::Deserialize(
kSkDrawable_Type, data, size, procs).release()));
}
Factory getFactory() const override { return nullptr; }
const char* getTypeName() const override { return nullptr; }
protected:
SkDrawable();
virtual SkRect onGetBounds() = 0;
virtual void onDraw(SkCanvas*) = 0;
virtual std::unique_ptr<GpuDrawHandler> onSnapGpuDrawHandler(GrBackendApi, const SkMatrix&,
const SkIRect& /*clipBounds*/,
const SkImageInfo&) {
return nullptr;
}
// TODO: Delete this once Android gets updated to take the clipBounds version above.
virtual std::unique_ptr<GpuDrawHandler> onSnapGpuDrawHandler(GrBackendApi, const SkMatrix&) {
return nullptr;
}
/**
* Default implementation calls onDraw() with a canvas that records into a picture. Subclasses
* may override if they have a more efficient way to return a picture for the current state
* of their drawable. Note: this picture must draw the same as what would be drawn from
* onDraw().
*/
virtual SkPicture* onNewPictureSnapshot();
private:
int32_t fGenerationID;
};
#endif