blob: c1642df81877f65a020c95232b6598344a24dece [file] [log] [blame]
/*
* Copyright 2020 Google LLC
*
* Use of this source code is governed by a BSD-style license that can be
* found in the LICENSE file.
*/
#ifndef SkMatrixProvider_DEFINED
#define SkMatrixProvider_DEFINED
#include "include/core/SkM44.h"
#include "include/core/SkMatrix.h"
/**
* All matrix providers report a flag: "localToDeviceHitsPixelCenters". This is confusing.
* It doesn't say anything about the actual matrix in the provider. Instead, it means: "is it safe
* to tweak sampling based on the contents of the matrix". In other words, does the device end of
* the local-to-device matrix actually map to pixels, AND are the local coordinates being fed to
* the shader produced by the inverse of that matrix? For a normal device, this is trivially true.
* The matrix may be updated via transforms, but when we draw (and the local coordinates come from
* rasterization of primitives against that device), we can know that the device coordinates will
* land on pixel centers.
*
* In a few places, the matrix provider is lying about how sampling "works". When we invoke a child
* from runtime effects, we give that child a matrix provider with an identity matrix. However --
* the coordinates being passed to that child are not the result of device -> local transformed
* coordinates. Runtime effects can generate coordinates arbitrarily - even though the provider has
* an identity matrix, we can't assume it's safe to (for example) convert linear -> nearest.
* Clip shaders are similar - they overwrite the local-to-device matrix (to match what it was when
* the clip shader was inserted). The CTM continues to change before drawing, though. In that case,
* the two matrices are not inverses, so the local coordinates may not land on texel centers in
* the clip shader.
*
* In cases where we need to inhibit filtering optimizations, use SkOverrideDeviceMatrixProvider.
*/
class SkMatrixProvider {
public:
SkMatrixProvider(const SkMatrix& localToDevice) : SkMatrixProvider(localToDevice, true) {}
SkMatrixProvider(const SkM44& localToDevice) : SkMatrixProvider(localToDevice, true) {}
// These should return the "same" matrix, as either a 3x3 or 4x4. Most sites in Skia still
// call localToDevice, and operate on SkMatrix.
const SkMatrix& localToDevice() const { return fLocalToDevice33; }
const SkM44& localToDevice44() const { return fLocalToDevice; }
bool localToDeviceHitsPixelCenters() const { return fHitsPixelCenters; }
protected:
SkMatrixProvider(const SkMatrix& localToDevice, bool hitsPixelCenters)
: fLocalToDevice(localToDevice)
, fLocalToDevice33(localToDevice)
, fHitsPixelCenters(hitsPixelCenters) {}
SkMatrixProvider(const SkM44& localToDevice, bool hitsPixelCenters)
: fLocalToDevice(localToDevice)
, fLocalToDevice33(localToDevice.asM33())
, fHitsPixelCenters(hitsPixelCenters) {}
private:
friend class SkBaseDevice;
SkM44 fLocalToDevice;
SkMatrix fLocalToDevice33; // Cached SkMatrix version of above, for legacy usage
const bool fHitsPixelCenters;
};
// Logically, this is equivalent to just making a new SkMatrixProvider, but we keep it distinct.
// This is for cases where there is an existing matrix provider, but we're replacing the
// local-to-device. In that scenario, we can't guarantee that we hit pixel centers anymore.
class SkOverrideDeviceMatrixProvider : public SkMatrixProvider {
public:
SkOverrideDeviceMatrixProvider(const SkMatrix& localToDevice)
: SkMatrixProvider(localToDevice, /*hitsPixelCenters=*/false) {}
};
class SkPostTranslateMatrixProvider : public SkMatrixProvider {
public:
SkPostTranslateMatrixProvider(const SkMatrixProvider& parent, SkScalar dx, SkScalar dy)
: SkMatrixProvider(SkM44::Translate(dx, dy) * parent.localToDevice44(),
parent.localToDeviceHitsPixelCenters()) {}
};
class SkPreConcatMatrixProvider : public SkMatrixProvider {
public:
SkPreConcatMatrixProvider(const SkMatrixProvider& parent, const SkMatrix& preMatrix)
: SkMatrixProvider(parent.localToDevice44() * SkM44(preMatrix),
parent.localToDeviceHitsPixelCenters()) {}
};
#endif