8724b46099
This reverts commit cd1416efbc
.
Reason for revert: speculative, to try to fix roll see gpu_tests.pixel_integration_test.PixelIntegrationTest.Pixel_GpuRasterization_ConcavePaths
Original change's description:
> Add support for semaphores to be inserted on GrContext flush
>
> This also moves the logic of inserting semaphores down into GrDrawingManager
> and finishFlush on GrGpu. With it being on finishFlush, there should be no
> issues when the DrawingManager starts respecting the proxy passed in assuming
> it always calls finishFlush at the end (which it should).
>
> Bug: skia:
> Change-Id: I925c2a289dcbbb9159b9120878af1d34f21a2dc7
> Reviewed-on: https://skia-review.googlesource.com/25641
> Reviewed-by: Brian Salomon <bsalomon@google.com>
> Commit-Queue: Greg Daniel <egdaniel@google.com>
TBR=egdaniel@google.com,bsalomon@google.com,robertphillips@google.com
Change-Id: I9c5b9cf8c060193e1861dbb8f0c10fb11dfb5249
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Bug: skia:
Reviewed-on: https://skia-review.googlesource.com/25980
Reviewed-by: Mike Reed <reed@google.com>
Commit-Queue: Mike Reed <reed@google.com>
372 lines
16 KiB
C++
372 lines
16 KiB
C++
/*
|
|
* Copyright 2012 Google Inc.
|
|
*
|
|
* Use of this source code is governed by a BSD-style license that can be
|
|
* found in the LICENSE file.
|
|
*/
|
|
|
|
#ifndef SkSurface_DEFINED
|
|
#define SkSurface_DEFINED
|
|
|
|
#include "SkRefCnt.h"
|
|
#include "SkImage.h"
|
|
#include "SkSurfaceProps.h"
|
|
|
|
class SkCanvas;
|
|
class SkPaint;
|
|
class GrBackendRenderTarget;
|
|
class GrBackendSemaphore;
|
|
class GrContext;
|
|
class GrRenderTarget;
|
|
|
|
/**
|
|
* SkSurface is responsible for managing the pixels that a canvas draws into. The pixels can be
|
|
* allocated either in CPU memory (a Raster surface) or on the GPU (a RenderTarget surface).
|
|
*
|
|
* SkSurface takes care of allocating a SkCanvas that will draw into the surface. Call
|
|
* surface->getCanvas() to use that canvas (but don't delete it, it is owned by the surface).
|
|
*
|
|
* SkSurface always has non-zero dimensions. If there is a request for a new surface, and either
|
|
* of the requested dimensions are zero, then NULL will be returned.
|
|
*/
|
|
class SK_API SkSurface : public SkRefCnt {
|
|
public:
|
|
/**
|
|
* Create a new surface, using the specified pixels/rowbytes as its
|
|
* backend.
|
|
*
|
|
* If the requested surface cannot be created, or the request is not a
|
|
* supported configuration, NULL will be returned.
|
|
*
|
|
* Callers are responsible for initialiazing the surface pixels.
|
|
*/
|
|
static sk_sp<SkSurface> MakeRasterDirect(const SkImageInfo&, void* pixels, size_t rowBytes,
|
|
const SkSurfaceProps* = nullptr);
|
|
|
|
/**
|
|
* The same as NewRasterDirect, but also accepts a call-back routine, which is invoked
|
|
* when the surface is deleted, and is passed the pixel memory and the specified context.
|
|
*/
|
|
static sk_sp<SkSurface> MakeRasterDirectReleaseProc(const SkImageInfo&, void* pixels, size_t rowBytes,
|
|
void (*releaseProc)(void* pixels, void* context),
|
|
void* context, const SkSurfaceProps* = nullptr);
|
|
|
|
/**
|
|
* Return a new surface, with the memory for the pixels automatically allocated and
|
|
* zero-initialized, but respecting the specified rowBytes. If rowBytes==0, then a default
|
|
* value will be chosen. If a non-zero rowBytes is specified, then any images snapped off of
|
|
* this surface (via makeImageSnapshot()) are guaranteed to have the same rowBytes.
|
|
*
|
|
* If the requested surface cannot be created, or the request is not a
|
|
* supported configuration, NULL will be returned.
|
|
*/
|
|
static sk_sp<SkSurface> MakeRaster(const SkImageInfo&, size_t rowBytes, const SkSurfaceProps*);
|
|
|
|
/**
|
|
* Allocate a new surface, automatically computing the rowBytes.
|
|
*/
|
|
static sk_sp<SkSurface> MakeRaster(const SkImageInfo& info,
|
|
const SkSurfaceProps* props = nullptr) {
|
|
return MakeRaster(info, 0, props);
|
|
}
|
|
|
|
/**
|
|
* Helper version of NewRaster. It creates a SkImageInfo with the
|
|
* specified width and height, and populates the rest of info to match
|
|
* pixels in SkPMColor format.
|
|
*/
|
|
static sk_sp<SkSurface> MakeRasterN32Premul(int width, int height,
|
|
const SkSurfaceProps* props = nullptr) {
|
|
return MakeRaster(SkImageInfo::MakeN32Premul(width, height), props);
|
|
}
|
|
|
|
/**
|
|
* Used to wrap a pre-existing backend 3D API texture as a SkSurface. Skia will not assume
|
|
* ownership of the texture and the client must ensure the texture is valid for the lifetime
|
|
* of the SkSurface. If sampleCnt > 0, then we will create an intermediate mssa surface which
|
|
* we will use for rendering. We then resolve into the passed in texture.
|
|
*/
|
|
static sk_sp<SkSurface> MakeFromBackendTexture(GrContext*, const GrBackendTexture&,
|
|
GrSurfaceOrigin origin, int sampleCnt,
|
|
sk_sp<SkColorSpace>, const SkSurfaceProps*);
|
|
|
|
/**
|
|
* Used to wrap a pre-existing 3D API rendering target as a SkSurface. Skia will not assume
|
|
* ownership of the render target and the client must ensure the render target is valid for the
|
|
* lifetime of the SkSurface.
|
|
*/
|
|
static sk_sp<SkSurface> MakeFromBackendRenderTarget(GrContext*,
|
|
const GrBackendRenderTargetDesc&,
|
|
sk_sp<SkColorSpace>,
|
|
const SkSurfaceProps*);
|
|
|
|
static sk_sp<SkSurface> MakeFromBackendRenderTarget(GrContext*,
|
|
const GrBackendRenderTarget&,
|
|
GrSurfaceOrigin origin,
|
|
sk_sp<SkColorSpace>,
|
|
const SkSurfaceProps*);
|
|
|
|
/**
|
|
* Used to wrap a pre-existing 3D API texture as a SkSurface. Skia will treat the texture as
|
|
* a rendering target only, but unlike NewFromBackendRenderTarget, Skia will manage and own
|
|
* the associated render target objects (but not the provided texture). The kRenderTarget flag
|
|
* must be set on GrBackendTextureDesc for this to succeed. Skia will not assume ownership
|
|
* of the texture and the client must ensure the texture is valid for the lifetime of the
|
|
* SkSurface.
|
|
*/
|
|
static sk_sp<SkSurface> MakeFromBackendTextureAsRenderTarget(GrContext*,
|
|
const GrBackendTexture&,
|
|
GrSurfaceOrigin origin,
|
|
int sampleCnt,
|
|
sk_sp<SkColorSpace>,
|
|
const SkSurfaceProps*);
|
|
|
|
/**
|
|
* Legacy version of the above factory, without color space support. This creates a "legacy"
|
|
* surface that operate without gamma correction or color management.
|
|
*/
|
|
static sk_sp<SkSurface> MakeFromBackendRenderTarget(GrContext* ctx,
|
|
const GrBackendRenderTargetDesc& desc,
|
|
const SkSurfaceProps* props) {
|
|
return MakeFromBackendRenderTarget(ctx, desc, nullptr, props);
|
|
}
|
|
|
|
|
|
/**
|
|
* Return a new surface whose contents will be drawn to an offscreen
|
|
* render target, allocated by the surface.
|
|
*/
|
|
static sk_sp<SkSurface> MakeRenderTarget(GrContext*, SkBudgeted, const SkImageInfo&,
|
|
int sampleCount, GrSurfaceOrigin,
|
|
const SkSurfaceProps*);
|
|
|
|
static sk_sp<SkSurface> MakeRenderTarget(GrContext* context, SkBudgeted budgeted,
|
|
const SkImageInfo& info, int sampleCount,
|
|
const SkSurfaceProps* props) {
|
|
return MakeRenderTarget(context, budgeted, info, sampleCount,
|
|
kBottomLeft_GrSurfaceOrigin, props);
|
|
}
|
|
|
|
static sk_sp<SkSurface> MakeRenderTarget(GrContext* gr, SkBudgeted b, const SkImageInfo& info) {
|
|
if (!info.width() || !info.height()) {
|
|
return nullptr;
|
|
}
|
|
return MakeRenderTarget(gr, b, info, 0, kBottomLeft_GrSurfaceOrigin, nullptr);
|
|
}
|
|
|
|
/**
|
|
* Returns a surface that stores no pixels. It can be drawn to via its canvas, but that
|
|
* canvas does not draw anything. Calling makeImageSnapshot() will return nullptr.
|
|
*/
|
|
static sk_sp<SkSurface> MakeNull(int width, int height);
|
|
|
|
int width() const { return fWidth; }
|
|
int height() const { return fHeight; }
|
|
|
|
/**
|
|
* Returns a unique non-zero, unique value identifying the content of this
|
|
* surface. Each time the content is changed changed, either by drawing
|
|
* into this surface, or explicitly calling notifyContentChanged()) this
|
|
* method will return a new value.
|
|
*
|
|
* If this surface is empty (i.e. has a zero-dimention), this will return
|
|
* 0.
|
|
*/
|
|
uint32_t generationID();
|
|
|
|
/**
|
|
* Modes that can be passed to notifyContentWillChange
|
|
*/
|
|
enum ContentChangeMode {
|
|
/**
|
|
* Use this mode if it is known that the upcoming content changes will
|
|
* clear or overwrite prior contents, thus making them discardable.
|
|
*/
|
|
kDiscard_ContentChangeMode,
|
|
/**
|
|
* Use this mode if prior surface contents need to be preserved or
|
|
* if in doubt.
|
|
*/
|
|
kRetain_ContentChangeMode,
|
|
};
|
|
|
|
/**
|
|
* Call this if the contents are about to change. This will (lazily) force a new
|
|
* value to be returned from generationID() when it is called next.
|
|
*
|
|
* CAN WE DEPRECATE THIS?
|
|
*/
|
|
void notifyContentWillChange(ContentChangeMode mode);
|
|
|
|
enum BackendHandleAccess {
|
|
kFlushRead_BackendHandleAccess, //!< caller may read from the backend object
|
|
kFlushWrite_BackendHandleAccess, //!< caller may write to the backend object
|
|
kDiscardWrite_BackendHandleAccess, //!< caller must over-write the entire backend object
|
|
};
|
|
|
|
/*
|
|
* These are legacy aliases which will be removed soon
|
|
*/
|
|
static const BackendHandleAccess kFlushRead_TextureHandleAccess =
|
|
kFlushRead_BackendHandleAccess;
|
|
static const BackendHandleAccess kFlushWrite_TextureHandleAccess =
|
|
kFlushWrite_BackendHandleAccess;
|
|
static const BackendHandleAccess kDiscardWrite_TextureHandleAccess =
|
|
kDiscardWrite_BackendHandleAccess;
|
|
|
|
|
|
/**
|
|
* Retrieves the backend API handle of the texture used by this surface, or 0 if the surface
|
|
* is not backed by a GPU texture.
|
|
*
|
|
* The returned texture-handle is only valid until the next draw-call into the surface,
|
|
* or the surface is deleted.
|
|
*/
|
|
GrBackendObject getTextureHandle(BackendHandleAccess);
|
|
|
|
/**
|
|
* Retrieves the backend API handle of the RenderTarget backing this surface. Callers must
|
|
* ensure this function returns 'true' or else the GrBackendObject will be invalid
|
|
*
|
|
* In OpenGL this will return the FramebufferObject ID.
|
|
*/
|
|
bool getRenderTargetHandle(GrBackendObject*, BackendHandleAccess);
|
|
|
|
/**
|
|
* Return a canvas that will draw into this surface. This will always
|
|
* return the same canvas for a given surface, and is manged/owned by the
|
|
* surface. It should not be used when its parent surface has gone out of
|
|
* scope.
|
|
*/
|
|
SkCanvas* getCanvas();
|
|
|
|
/**
|
|
* Return a new surface that is "compatible" with this one, in that it will
|
|
* efficiently be able to be drawn into this surface. Typical calling
|
|
* pattern:
|
|
*
|
|
* SkSurface* A = SkSurface::New...();
|
|
* SkCanvas* canvasA = surfaceA->newCanvas();
|
|
* ...
|
|
* SkSurface* surfaceB = surfaceA->newSurface(...);
|
|
* SkCanvas* canvasB = surfaceB->newCanvas();
|
|
* ... // draw using canvasB
|
|
* canvasA->drawSurface(surfaceB); // <--- this will always be optimal!
|
|
*/
|
|
sk_sp<SkSurface> makeSurface(const SkImageInfo&);
|
|
|
|
/**
|
|
* Returns an image of the current state of the surface pixels up to this
|
|
* point. Subsequent changes to the surface (by drawing into its canvas)
|
|
* will not be reflected in this image. For the GPU-backend, the budgeting
|
|
* decision for the snapped image will match that of the surface.
|
|
*/
|
|
sk_sp<SkImage> makeImageSnapshot();
|
|
|
|
/**
|
|
* Though the caller could get a snapshot image explicitly, and draw that,
|
|
* it seems that directly drawing a surface into another canvas might be
|
|
* a common pattern, and that we could possibly be more efficient, since
|
|
* we'd know that the "snapshot" need only live until we've handed it off
|
|
* to the canvas.
|
|
*/
|
|
void draw(SkCanvas*, SkScalar x, SkScalar y, const SkPaint*);
|
|
|
|
/**
|
|
* If the surface has direct access to its pixels (i.e. they are in local
|
|
* RAM) return true, and if not null, set the pixmap parameter to point to the information
|
|
* about the surface's pixels. The pixel address in the pixmap is only valid while
|
|
* the surface object is in scope, and no API call is made on the surface
|
|
* or its canvas.
|
|
*
|
|
* On failure, returns false and the pixmap parameter is ignored.
|
|
*/
|
|
bool peekPixels(SkPixmap*);
|
|
|
|
/**
|
|
* Copy the pixels from the surface into the specified pixmap,
|
|
* converting them into the pixmap's format. The surface pixels are read
|
|
* starting at the specified (srcX,srcY) location.
|
|
*
|
|
* The pixmap and (srcX,srcY) offset specifies a source rectangle
|
|
*
|
|
* srcR.setXYWH(srcX, srcY, pixmap.width(), pixmap.height());
|
|
*
|
|
* srcR is intersected with the bounds of the base-layer. If this intersection is not empty,
|
|
* then we have two sets of pixels (of equal size). Replace the dst pixels with the
|
|
* corresponding src pixels, performing any colortype/alphatype transformations needed
|
|
* (in the case where the src and dst have different colortypes or alphatypes).
|
|
*
|
|
* This call can fail, returning false, for several reasons:
|
|
* - If srcR does not intersect the surface bounds.
|
|
* - If the requested colortype/alphatype cannot be converted from the surface's types.
|
|
*/
|
|
bool readPixels(const SkPixmap& dst, int srcX, int srcY);
|
|
bool readPixels(const SkImageInfo& dstInfo, void* dstPixels, size_t dstRowBytes,
|
|
int srcX, int srcY);
|
|
bool readPixels(const SkBitmap& dst, int srcX, int srcY);
|
|
|
|
const SkSurfaceProps& props() const { return fProps; }
|
|
|
|
/**
|
|
* Issue any pending surface IO to the current backend 3D API and resolve any surface MSAA.
|
|
*
|
|
* The flush calls below are the new preferred way to flush calls to a surface, and this call
|
|
* will eventually be removed.
|
|
*/
|
|
void prepareForExternalIO();
|
|
|
|
/**
|
|
* Issue any pending surface IO to the current backend 3D API
|
|
*/
|
|
void flush();
|
|
|
|
/**
|
|
* Issue any pending surface IO to the current backend 3D API. After issuing all commands,
|
|
* numSemaphore semaphores will be signaled by the gpu. The client passes in an array of
|
|
* numSemaphores GrBackendSemaphores. In general these GrBackendSemaphore's can be either
|
|
* initialized or not. If they are initialized, the backend uses the passed in semaphore.
|
|
* If it is not initialized, a new semaphore is created and the GrBackendSemaphore object
|
|
* is initialized with that semaphore.
|
|
*
|
|
* If the backend API is OpenGL only uninitialized GrBackendSemaphores are supported.
|
|
* If the backend API is Vulkan either initialized or unitialized semaphores are supported.
|
|
* If unitialized, the semaphores which are created will be valid for use only with the VkDevice
|
|
* with which they were created.
|
|
*
|
|
* If this call returns false, the GPU backend will not have created or added any semaphores to
|
|
* signal. Thus the array of semaphores will remain uninitialized. However, any pending surface
|
|
* IO will still be flush.
|
|
*/
|
|
bool flushAndSignalSemaphores(int numSemaphores, GrBackendSemaphore* signalSemaphores);
|
|
|
|
/**
|
|
* Inserts a list of GPU semaphores that the current backend 3D API must wait on before
|
|
* executing any more commands on the GPU for this surface. Skia will take ownership of the
|
|
* underlying semaphores and delete them once they have been signaled and waited on.
|
|
*
|
|
* If this call returns false, then the GPU backend will not wait on any passed in semaphores,
|
|
* and the client will still own the semaphores.
|
|
*/
|
|
bool wait(int numSemaphores, const GrBackendSemaphore* waitSemaphores);
|
|
|
|
protected:
|
|
SkSurface(int width, int height, const SkSurfaceProps*);
|
|
SkSurface(const SkImageInfo&, const SkSurfaceProps*);
|
|
|
|
// called by subclass if their contents have changed
|
|
void dirtyGenerationID() {
|
|
fGenerationID = 0;
|
|
}
|
|
|
|
private:
|
|
const SkSurfaceProps fProps;
|
|
const int fWidth;
|
|
const int fHeight;
|
|
uint32_t fGenerationID;
|
|
|
|
typedef SkRefCnt INHERITED;
|
|
};
|
|
|
|
#endif
|