skia2/include/gpu/GrXferProcessor.h

394 lines
15 KiB
C
Raw Normal View History

/*
* 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 GrXferProcessor_DEFINED
#define GrXferProcessor_DEFINED
#include "GrBlend.h"
#include "GrColor.h"
#include "GrProcessor.h"
#include "GrTexture.h"
#include "GrTypes.h"
#include "SkXfermode.h"
class GrShaderCaps;
class GrGLSLCaps;
class GrGLSLXferProcessor;
class GrProcOptInfo;
struct GrPipelineOptimizations;
/**
* Barriers for blending. When a shader reads the dst directly, an Xfer barrier is sometimes
* required after a pixel has been written, before it can be safely read again.
*/
enum GrXferBarrierType {
kNone_GrXferBarrierType = 0, //<! No barrier is required
kTexture_GrXferBarrierType, //<! Required when a shader reads and renders to the same texture.
kBlend_GrXferBarrierType, //<! Required by certain blend extensions.
};
/** Should be able to treat kNone as false in boolean expressions */
GR_STATIC_ASSERT(SkToBool(kNone_GrXferBarrierType) == false);
/**
* GrXferProcessor is responsible for implementing the xfer mode that blends the src color and dst
* color, and for applying any coverage. It does this by emitting fragment shader code and
* controlling the fixed-function blend state. When dual-source blending is available, it may also
* write a seconday fragment shader output color. GrXferProcessor has two modes of operation:
*
* Dst read: When allowed by the backend API, or when supplied a texture of the destination, the
* GrXferProcessor may read the destination color. While operating in this mode, the subclass only
* provides shader code that blends the src and dst colors, and the base class applies coverage.
*
* No dst read: When not performing a dst read, the subclass is given full control of the fixed-
* function blend state and/or secondary output, and is responsible to apply coverage on its own.
*
* A GrXferProcessor is never installed directly into our draw state, but instead is created from a
* GrXPFactory once we have finalized the state of our draw.
*/
class GrXferProcessor : public GrProcessor {
public:
/**
* A texture that contains the dst pixel values and an integer coord offset from device space
* to the space of the texture. Depending on GPU capabilities a DstTexture may be used by a
* GrXferProcessor for blending in the fragment shader.
*/
class DstTexture {
public:
DstTexture() { fOffset.set(0, 0); }
DstTexture(const DstTexture& other) {
*this = other;
}
DstTexture(GrTexture* texture, const SkIPoint& offset)
: fTexture(SkSafeRef(texture))
, fOffset(offset) {
}
DstTexture& operator=(const DstTexture& other) {
fTexture.reset(SkSafeRef(other.fTexture.get()));
fOffset = other.fOffset;
return *this;
}
const SkIPoint& offset() const { return fOffset; }
void setOffset(const SkIPoint& offset) { fOffset = offset; }
void setOffset(int ox, int oy) { fOffset.set(ox, oy); }
GrTexture* texture() const { return fTexture.get(); }
GrTexture* setTexture(GrTexture* texture) {
fTexture.reset(SkSafeRef(texture));
return texture;
}
private:
SkAutoTUnref<GrTexture> fTexture;
SkIPoint fOffset;
};
/**
* Sets a unique key on the GrProcessorKeyBuilder calls onGetGLSLProcessorKey(...) to get the
* specific subclass's key.
*/
void getGLSLProcessorKey(const GrGLSLCaps& caps, GrProcessorKeyBuilder* b) const;
/** Returns a new instance of the appropriate *GL* implementation class
for the given GrXferProcessor; caller is responsible for deleting
the object. */
virtual GrGLSLXferProcessor* createGLSLInstance() const = 0;
/**
* Optimizations for blending / coverage that an OptDrawState should apply to itself.
*/
enum OptFlags {
/**
* The draw can be skipped completely.
*/
kSkipDraw_OptFlag = 0x1,
/**
* GrXferProcessor will ignore color, thus no need to provide
*/
kIgnoreColor_OptFlag = 0x2,
/**
* GrXferProcessor will ignore coverage, thus no need to provide
*/
kIgnoreCoverage_OptFlag = 0x4,
/**
* Clear color stages and override input color to that returned by getOptimizations
*/
kOverrideColor_OptFlag = 0x8,
Revert of Implement Porter Duff XP with a blend table (patchset #12 id:220001 of https://codereview.chromium.org/1124373002/) Reason for revert: Blocking DEPS roll into Chromium. Crashing virtual/gpu/fast/canvas/canvas-composite-*.html tests with the assert ../../third_party/skia/src/gpu/gl/builders/GrGLFragmentShaderBuilder.cpp:281: failed assertion "k110_GrGLSLGeneration != gpu->glslGeneration() || fOutputs.empty()" Original issue's description: > Implement Porter Duff XP with a blend table > > Removes the runtime logic used by PorterDuffXferProcessor to decide > blend coeffs and shader outputs, and instead uses a compile-time > constant table of pre-selected blend formulas. > > Introduces a new blend strategy for srcCoeff=0 that can apply coverage > with a reverse subtract blend equation instead of dual source > blending. > > Adds new macros in GrBlend.h to analyze blend formulas both runtime. > > Removes kSetCoverageDrawing_OptFlag and GrSimplifyBlend as they are no > longer used. > > Adds a GM that verifies all xfermodes, including arithmetic, with the > color/coverage invariants used by Porter Duff. > > Adds a unit test that verifies each Porter Duff formula with every > color/coverage invariant. > > Major changes: > > * Uses a reverse subtract blend equation for coverage when srcCoeff=0 > (clear, dst-out [Sa=1], dst-in, modulate). Platforms that don't > support dual source blending no longer require a dst copy for > dst-in and modulate. > > * Sets BlendInfo::fWriteColor to false when the blend does not modify > the dst. GrGLGpu will now use glColorMask instead of blending for > these modes (dst, dst-in [Sa=1], modulate ignored for [Sc=1]). > > * Converts all SA blend coeffs to One for opaque inputs, and ISA to > Zero if there is also no coverage. (We keep ISA around when there > is coverage because we use it to tweak alpha for coverage.) > > * Abandons solid white optimizations for the sake of simplicity > (screen was the only mode that previous had solid white opts). > > Minor differences: > > * Inconsequential differences in opt flags (e.g. we now return > kCanTweakAlphaForCoverage_OptFlag even when there is no coverage). > > * Src coeffs when the shader outputs 0. > > * IS2C vs IS2A when the secondary output is scalar. > > BUG=skia: > > Committed: https://skia.googlesource.com/skia/+/9a70920db22b6309c671f8e5d519bb95570e4414 TBR=egdaniel@google.com,bsalomon@google.com,cdalton@nvidia.com NOPRESUBMIT=true NOTREECHECKS=true NOTRY=true BUG=skia: Review URL: https://codereview.chromium.org/1153993002
2015-05-23 00:55:26 +00:00
/**
* Can tweak alpha for coverage. Currently this flag should only be used by a batch
*/
kCanTweakAlphaForCoverage_OptFlag = 0x20,
};
static const OptFlags kNone_OptFlags = (OptFlags)0;
GR_DECL_BITFIELD_OPS_FRIENDS(OptFlags);
/**
* Determines which optimizations (as described by the ptFlags above) can be performed by
* the draw with this xfer processor. If this function is called, the xfer processor may change
* its state to reflected the given blend optimizations. If the XP needs to see a specific input
* color to blend correctly, it will set the OverrideColor flag and the output parameter
* overrideColor will be the required value that should be passed into the XP.
* A caller who calls this function on a XP is required to honor the returned OptFlags
* and color values for its draw.
*/
OptFlags getOptimizations(const GrPipelineOptimizations& optimizations,
bool doesStencilWrite,
GrColor* overrideColor,
const GrCaps& caps) const;
/**
* Returns whether this XP will require an Xfer barrier on the given rt. If true, outBarrierType
* is updated to contain the type of barrier needed.
*/
GrXferBarrierType xferBarrierType(const GrRenderTarget* rt, const GrCaps& caps) const;
struct BlendInfo {
void reset() {
fEquation = kAdd_GrBlendEquation;
fSrcBlend = kOne_GrBlendCoeff;
fDstBlend = kZero_GrBlendCoeff;
fBlendConstant = 0;
fWriteColor = true;
}
SkDEBUGCODE(SkString dump() const;)
GrBlendEquation fEquation;
GrBlendCoeff fSrcBlend;
GrBlendCoeff fDstBlend;
GrColor fBlendConstant;
bool fWriteColor;
};
void getBlendInfo(BlendInfo* blendInfo) const;
bool willReadDstColor() const { return fWillReadDstColor; }
/**
* Returns the texture to be used as the destination when reading the dst in the fragment
* shader. If the returned texture is NULL then the XP is either not reading the dst or we have
* extentions that support framebuffer fetching and thus don't need a copy of the dst texture.
*/
const GrTexture* getDstTexture() const { return fDstTexture.getTexture(); }
/**
* Returns the offset in device coords to use when accessing the dst texture to get the dst
* pixel color in the shader. This value is only valid if getDstTexture() != NULL.
*/
const SkIPoint& dstTextureOffset() const {
SkASSERT(this->getDstTexture());
return fDstTextureOffset;
}
/**
* If we are performing a dst read, returns whether the base class will use mixed samples to
* antialias the shader's final output. If not doing a dst read, the subclass is responsible
* for antialiasing and this returns false.
*/
bool dstReadUsesMixedSamples() const { return fDstReadUsesMixedSamples; }
/**
* Returns whether or not this xferProcossor will set a secondary output to be used with dual
* source blending.
*/
bool hasSecondaryOutput() const;
/** Returns true if this and other processor conservatively draw identically. It can only return
true when the two processor are of the same subclass (i.e. they return the same object from
from getFactory()).
A return value of true from isEqual() should not be used to test whether the processor would
generate the same shader code. To test for identical code generation use getGLSLProcessorKey
*/
bool isEqual(const GrXferProcessor& that) const {
if (this->classID() != that.classID()) {
return false;
}
if (this->fWillReadDstColor != that.fWillReadDstColor) {
return false;
}
if (this->fDstTexture.getTexture() != that.fDstTexture.getTexture()) {
return false;
}
if (this->fDstTextureOffset != that.fDstTextureOffset) {
return false;
}
if (this->fDstReadUsesMixedSamples != that.fDstReadUsesMixedSamples) {
return false;
}
return this->onIsEqual(that);
}
protected:
GrXferProcessor();
GrXferProcessor(const DstTexture*, bool willReadDstColor, bool hasMixedSamples);
private:
void notifyRefCntIsZero() const final {}
virtual OptFlags onGetOptimizations(const GrPipelineOptimizations& optimizations,
bool doesStencilWrite,
GrColor* overrideColor,
const GrCaps& caps) const = 0;
/**
* Sets a unique key on the GrProcessorKeyBuilder that is directly associated with this xfer
* processor's GL backend implementation.
*/
virtual void onGetGLSLProcessorKey(const GrGLSLCaps& caps,
GrProcessorKeyBuilder* b) const = 0;
/**
* Determines the type of barrier (if any) required by the subclass. Note that the possibility
* that a kTexture type barrier is required is handled by the base class and need not be
* considered by subclass overrides of this function.
*/
virtual GrXferBarrierType onXferBarrier(const GrRenderTarget*, const GrCaps&) const {
return kNone_GrXferBarrierType;
}
/**
* If we are not performing a dst read, returns whether the subclass will set a secondary
* output. When using dst reads, the base class controls the secondary output and this method
* will not be called.
*/
virtual bool onHasSecondaryOutput() const { return false; }
/**
* If we are not performing a dst read, retrieves the fixed-function blend state required by the
* subclass. When using dst reads, the base class controls the fixed-function blend state and
* this method will not be called. The BlendInfo struct comes initialized to "no blending".
*/
virtual void onGetBlendInfo(BlendInfo*) const {}
virtual bool onIsEqual(const GrXferProcessor&) const = 0;
bool fWillReadDstColor;
bool fDstReadUsesMixedSamples;
SkIPoint fDstTextureOffset;
GrTextureAccess fDstTexture;
typedef GrFragmentProcessor INHERITED;
};
GR_MAKE_BITFIELD_OPS(GrXferProcessor::OptFlags);
///////////////////////////////////////////////////////////////////////////////
/**
* We install a GrXPFactory (XPF) early on in the pipeline before all the final draw information is
* known (e.g. whether there is fractional pixel coverage, will coverage be 1 or 4 channel, is the
* draw opaque, etc.). Once the state of the draw is finalized, we use the XPF along with all the
* draw information to create a GrXferProcessor (XP) which can implement the desired blending for
* the draw.
*
* Before the XP is created, the XPF is able to answer queries about what functionality the XPs it
* creates will have. For example, can it create an XP that supports RGB coverage or will the XP
* blend with the destination color.
*/
class GrXPFactory : public SkRefCnt {
public:
typedef GrXferProcessor::DstTexture DstTexture;
GrXferProcessor* createXferProcessor(const GrPipelineOptimizations& optimizations,
bool hasMixedSamples,
const DstTexture*,
const GrCaps& caps) const;
/**
* Known color information after blending, but before accounting for any coverage.
*/
struct InvariantBlendedColor {
bool fWillBlendWithDst;
GrColor fKnownColor;
GrColorComponentFlags fKnownColorFlags;
};
/**
* Returns information about the output color, produced by XPs from this factory, that will be
* known after blending. Note that we can conflate coverage and color, so the actual values
* written to pixels with partial coverage may not always seem consistent with the invariant
* information returned by this function.
*/
virtual void getInvariantBlendedColor(const GrProcOptInfo& colorPOI,
InvariantBlendedColor*) const = 0;
bool willNeedDstTexture(const GrCaps& caps, const GrPipelineOptimizations& optimizations,
bool hasMixedSamples) const;
bool isEqual(const GrXPFactory& that) const {
if (this->classID() != that.classID()) {
return false;
}
return this->onIsEqual(that);
}
/**
* Helper for down-casting to a GrXPFactory subclass
*/
template <typename T> const T& cast() const { return *static_cast<const T*>(this); }
uint32_t classID() const { SkASSERT(kIllegalXPFClassID != fClassID); return fClassID; }
protected:
GrXPFactory() : fClassID(kIllegalXPFClassID) {}
template <typename XPF_SUBCLASS> void initClassID() {
static uint32_t kClassID = GenClassID();
fClassID = kClassID;
}
uint32_t fClassID;
private:
virtual GrXferProcessor* onCreateXferProcessor(const GrCaps& caps,
const GrPipelineOptimizations& optimizations,
bool hasMixedSamples,
const DstTexture*) const = 0;
virtual bool onIsEqual(const GrXPFactory&) const = 0;
bool willReadDstColor(const GrCaps& caps,
const GrPipelineOptimizations& optimizations,
bool hasMixedSamples) const;
/**
* Returns true if the XP generated by this factory will explicitly read dst in the fragment
* shader.
*/
virtual bool onWillReadDstColor(const GrCaps& caps,
const GrPipelineOptimizations& optimizations,
bool hasMixedSamples) const = 0;
static uint32_t GenClassID() {
// fCurrXPFactoryID has been initialized to kIllegalXPFactoryID. The
// atomic inc returns the old value not the incremented value. So we add
// 1 to the returned value.
uint32_t id = static_cast<uint32_t>(sk_atomic_inc(&gCurrXPFClassID)) + 1;
if (!id) {
SkFAIL("This should never wrap as it should only be called once for each GrXPFactory "
"subclass.");
}
return id;
}
enum {
kIllegalXPFClassID = 0,
};
static int32_t gCurrXPFClassID;
typedef GrProgramElement INHERITED;
};
#endif