skia2/tests/GrClipStackTest.cpp

2159 lines
99 KiB
C++
Raw Normal View History

New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
/*
* Copyright 2020 Google LLC
*
* Use of this source code is governed by a BSD-style license that can be
* found in the LICENSE file.
*/
#include "src/gpu/v1/ClipStack.h"
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
#include "tests/Test.h"
#include "include/core/SkPath.h"
#include "include/core/SkRRect.h"
#include "include/core/SkRect.h"
#include "include/core/SkRegion.h"
#include "include/core/SkShader.h"
#include "include/gpu/GrDirectContext.h"
#include "src/core/SkMatrixProvider.h"
#include "src/core/SkRRectPriv.h"
#include "src/core/SkRectPriv.h"
#include "src/gpu/GrDirectContextPriv.h"
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
#include "src/gpu/GrProxyProvider.h"
#include "src/gpu/ops/GrDrawOp.h"
#include "src/gpu/v1/SurfaceDrawContext_v1.h"
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
namespace {
class TestCaseBuilder;
class ElementsBuilder;
enum class SavePolicy {
kNever,
kAtStart,
kAtEnd,
kBetweenEveryOp
};
// TODO: We could add a RestorePolicy enum that tests different places to restore, but that would
// make defining the test expectations and order independence more cumbersome.
class TestCase {
public:
using ClipStack = skgpu::v1::ClipStack;
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
// Provides fluent API to describe actual clip commands and expected clip elements:
// TestCase test = TestCase::Build("example", deviceBounds)
// .actual().rect(r, GrAA::kYes, SkClipOp::kIntersect)
// .localToDevice(matrix)
// .nonAA()
// .difference()
// .path(p1)
// .path(p2)
// .finishElements()
// .expectedState(kDeviceRect)
// .expectedBounds(r.roundOut())
// .expect().rect(r, GrAA::kYes, SkClipOp::kIntersect)
// .finishElements()
// .finishTest();
static TestCaseBuilder Build(const char* name, const SkIRect& deviceBounds);
void run(const std::vector<int>& order, SavePolicy policy, skiatest::Reporter* reporter) const;
const SkIRect& deviceBounds() const { return fDeviceBounds; }
ClipStack::ClipState expectedState() const { return fExpectedState; }
const std::vector<ClipStack::Element>& initialElements() const { return fElements; }
const std::vector<ClipStack::Element>& expectedElements() const { return fExpectedElements; }
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
private:
friend class TestCaseBuilder;
TestCase(SkString name,
const SkIRect& deviceBounds,
ClipStack::ClipState expectedState,
std::vector<ClipStack::Element> actual,
std::vector<ClipStack::Element> expected)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
: fName(name)
, fElements(std::move(actual))
, fDeviceBounds(deviceBounds)
, fExpectedElements(std::move(expected))
, fExpectedState(expectedState) {}
SkString getTestName(const std::vector<int>& order, SavePolicy policy) const;
// This may be tighter than ClipStack::getConservativeBounds() because this always accounts
// for difference ops, whereas ClipStack only sometimes can subtract the inner bounds for a
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
// difference op.
std::pair<SkIRect, bool> getOptimalBounds() const;
SkString fName;
// The input shapes+state to ClipStack
std::vector<ClipStack::Element> fElements;
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
SkIRect fDeviceBounds;
// The expected output of iterating over the ClipStack after all fElements are added, although
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
// order is not important
std::vector<ClipStack::Element> fExpectedElements;
ClipStack::ClipState fExpectedState;
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
};
class ElementsBuilder {
public:
using ClipStack = skgpu::v1::ClipStack;
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
// Update the default matrix, aa, and op state for elements that are added.
ElementsBuilder& localToDevice(const SkMatrix& m) { fLocalToDevice = m; return *this; }
ElementsBuilder& aa() { fAA = GrAA::kYes; return *this; }
ElementsBuilder& nonAA() { fAA = GrAA::kNo; return *this; }
ElementsBuilder& intersect() { fOp = SkClipOp::kIntersect; return *this; }
ElementsBuilder& difference() { fOp = SkClipOp::kDifference; return *this; }
// Add rect, rrect, or paths to the list of elements, possibly overriding the last set
// matrix, aa, and op state.
ElementsBuilder& rect(const SkRect& rect) {
return this->rect(rect, fLocalToDevice, fAA, fOp);
}
ElementsBuilder& rect(const SkRect& rect, GrAA aa, SkClipOp op) {
return this->rect(rect, fLocalToDevice, aa, op);
}
ElementsBuilder& rect(const SkRect& rect, const SkMatrix& m, GrAA aa, SkClipOp op) {
fElements->push_back({GrShape(rect), m, op, aa});
return *this;
}
ElementsBuilder& rrect(const SkRRect& rrect) {
return this->rrect(rrect, fLocalToDevice, fAA, fOp);
}
ElementsBuilder& rrect(const SkRRect& rrect, GrAA aa, SkClipOp op) {
return this->rrect(rrect, fLocalToDevice, aa, op);
}
ElementsBuilder& rrect(const SkRRect& rrect, const SkMatrix& m, GrAA aa, SkClipOp op) {
fElements->push_back({GrShape(rrect), m, op, aa});
return *this;
}
ElementsBuilder& path(const SkPath& path) {
return this->path(path, fLocalToDevice, fAA, fOp);
}
ElementsBuilder& path(const SkPath& path, GrAA aa, SkClipOp op) {
return this->path(path, fLocalToDevice, aa, op);
}
ElementsBuilder& path(const SkPath& path, const SkMatrix& m, GrAA aa, SkClipOp op) {
fElements->push_back({GrShape(path), m, op, aa});
return *this;
}
// Finish and return the original test case builder
TestCaseBuilder& finishElements() {
return *fBuilder;
}
private:
friend class TestCaseBuilder;
ElementsBuilder(TestCaseBuilder* builder, std::vector<ClipStack::Element>* elements)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
: fBuilder(builder)
, fElements(elements) {}
SkMatrix fLocalToDevice = SkMatrix::I();
GrAA fAA = GrAA::kNo;
SkClipOp fOp = SkClipOp::kIntersect;
TestCaseBuilder* fBuilder;
std::vector<ClipStack::Element>* fElements;
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
};
class TestCaseBuilder {
public:
using ClipStack = skgpu::v1::ClipStack;
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
ElementsBuilder actual() { return ElementsBuilder(this, &fActualElements); }
ElementsBuilder expect() { return ElementsBuilder(this, &fExpectedElements); }
TestCaseBuilder& expectActual() {
fExpectedElements = fActualElements;
return *this;
}
TestCaseBuilder& state(ClipStack::ClipState state) {
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
fExpectedState = state;
return *this;
}
TestCase finishTest() {
TestCase test(fName, fDeviceBounds, fExpectedState,
std::move(fActualElements), std::move(fExpectedElements));
fExpectedState = ClipStack::ClipState::kWideOpen;
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
return test;
}
private:
friend class TestCase;
explicit TestCaseBuilder(const char* name, const SkIRect& deviceBounds)
: fName(name)
, fDeviceBounds(deviceBounds)
, fExpectedState(ClipStack::ClipState::kWideOpen) {}
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
SkString fName;
SkIRect fDeviceBounds;
ClipStack::ClipState fExpectedState;
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
std::vector<ClipStack::Element> fActualElements;
std::vector<ClipStack::Element> fExpectedElements;
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
};
TestCaseBuilder TestCase::Build(const char* name, const SkIRect& deviceBounds) {
return TestCaseBuilder(name, deviceBounds);
}
SkString TestCase::getTestName(const std::vector<int>& order, SavePolicy policy) const {
SkString name = fName;
SkString policyName;
switch(policy) {
case SavePolicy::kNever:
policyName = "never";
break;
case SavePolicy::kAtStart:
policyName = "start";
break;
case SavePolicy::kAtEnd:
policyName = "end";
break;
case SavePolicy::kBetweenEveryOp:
policyName = "between";
break;
}
name.appendf("(save %s, order [", policyName.c_str());
for (size_t i = 0; i < order.size(); ++i) {
if (i > 0) {
name.append(",");
}
name.appendf("%d", order[i]);
}
name.append("])");
return name;
}
std::pair<SkIRect, bool> TestCase::getOptimalBounds() const {
if (fExpectedState == ClipStack::ClipState::kEmpty) {
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
return {SkIRect::MakeEmpty(), true};
}
bool expectOptimal = true;
SkRegion region(fDeviceBounds);
for (const ClipStack::Element& e : fExpectedElements) {
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
bool intersect = (e.fOp == SkClipOp::kIntersect && !e.fShape.inverted()) ||
(e.fOp == SkClipOp::kDifference && e.fShape.inverted());
SkIRect elementBounds;
SkRegion::Op op;
if (intersect) {
op = SkRegion::kIntersect_Op;
expectOptimal &= e.fLocalToDevice.isIdentity();
elementBounds = GrClip::GetPixelIBounds(e.fLocalToDevice.mapRect(e.fShape.bounds()),
e.fAA, GrClip::BoundsType::kExterior);
} else {
op = SkRegion::kDifference_Op;
expectOptimal = false;
if (e.fShape.isRect() && e.fLocalToDevice.isIdentity()) {
elementBounds = GrClip::GetPixelIBounds(e.fShape.rect(), e.fAA,
GrClip::BoundsType::kInterior);
} else if (e.fShape.isRRect() && e.fLocalToDevice.isIdentity()) {
elementBounds = GrClip::GetPixelIBounds(SkRRectPriv::InnerBounds(e.fShape.rrect()),
e.fAA, GrClip::BoundsType::kInterior);
} else {
elementBounds = SkIRect::MakeEmpty();
}
}
region.op(SkRegion(elementBounds), op);
}
return {region.getBounds(), expectOptimal};
}
static bool compare_elements(const skgpu::v1::ClipStack::Element& a,
const skgpu::v1::ClipStack::Element& b) {
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
if (a.fAA != b.fAA || a.fOp != b.fOp || a.fLocalToDevice != b.fLocalToDevice ||
a.fShape.type() != b.fShape.type()) {
return false;
}
switch(a.fShape.type()) {
case GrShape::Type::kRect:
return a.fShape.rect() == b.fShape.rect();
case GrShape::Type::kRRect:
return a.fShape.rrect() == b.fShape.rrect();
case GrShape::Type::kPath:
// A path's points are never transformed, the only modification is fill type which does
// not change the generation ID. For convex polygons, we check == so that more complex
// test cases can be evaluated.
return a.fShape.path().getGenerationID() == b.fShape.path().getGenerationID() ||
(a.fShape.convex() &&
a.fShape.segmentMask() == SkPathSegmentMask::kLine_SkPathSegmentMask &&
a.fShape.path() == b.fShape.path());
default:
SkDEBUGFAIL("Shape type not handled by test case yet.");
return false;
}
}
void TestCase::run(const std::vector<int>& order,
SavePolicy policy,
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
skiatest::Reporter* reporter) const {
SkASSERT(fElements.size() == order.size());
SkMatrixProvider matrixProvider(SkMatrix::I());
ClipStack cs(fDeviceBounds, &matrixProvider, false);
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
if (policy == SavePolicy::kAtStart) {
cs.save();
}
for (int i : order) {
if (policy == SavePolicy::kBetweenEveryOp) {
cs.save();
}
const ClipStack::Element& e = fElements[i];
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
switch(e.fShape.type()) {
case GrShape::Type::kRect:
cs.clipRect(e.fLocalToDevice, e.fShape.rect(), e.fAA, e.fOp);
break;
case GrShape::Type::kRRect:
cs.clipRRect(e.fLocalToDevice, e.fShape.rrect(), e.fAA, e.fOp);
break;
case GrShape::Type::kPath:
cs.clipPath(e.fLocalToDevice, e.fShape.path(), e.fAA, e.fOp);
break;
default:
SkDEBUGFAIL("Shape type not handled by test case yet.");
}
}
if (policy == SavePolicy::kAtEnd) {
cs.save();
}
// Now validate
SkString name = this->getTestName(order, policy);
REPORTER_ASSERT(reporter, cs.clipState() == fExpectedState,
"%s, clip state expected %d, actual %d",
name.c_str(), (int) fExpectedState, (int) cs.clipState());
SkIRect actualBounds = cs.getConservativeBounds();
SkIRect optimalBounds;
bool expectOptimal;
std::tie(optimalBounds, expectOptimal) = this->getOptimalBounds();
if (expectOptimal) {
REPORTER_ASSERT(reporter, actualBounds == optimalBounds,
"%s, bounds expected [%d %d %d %d], actual [%d %d %d %d]",
name.c_str(), optimalBounds.fLeft, optimalBounds.fTop,
optimalBounds.fRight, optimalBounds.fBottom,
actualBounds.fLeft, actualBounds.fTop,
actualBounds.fRight, actualBounds.fBottom);
} else {
REPORTER_ASSERT(reporter, actualBounds.contains(optimalBounds),
"%s, bounds are not conservative, optimal [%d %d %d %d], actual [%d %d %d %d]",
name.c_str(), optimalBounds.fLeft, optimalBounds.fTop,
optimalBounds.fRight, optimalBounds.fBottom,
actualBounds.fLeft, actualBounds.fTop,
actualBounds.fRight, actualBounds.fBottom);
}
size_t matchedElements = 0;
for (const ClipStack::Element& a : cs) {
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
bool found = false;
for (const ClipStack::Element& e : fExpectedElements) {
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
if (compare_elements(a, e)) {
// shouldn't match multiple expected elements or it's a bad test case
SkASSERT(!found);
found = true;
}
}
REPORTER_ASSERT(reporter, found,
"%s, unexpected clip element in stack: shape %d, aa %d, op %d",
name.c_str(), (int) a.fShape.type(), (int) a.fAA, (int) a.fOp);
matchedElements += found ? 1 : 0;
}
REPORTER_ASSERT(reporter, matchedElements == fExpectedElements.size(),
"%s, did not match all expected elements: expected %zu but matched only %zu",
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
name.c_str(), fExpectedElements.size(), matchedElements);
// Validate restoration behavior
if (policy == SavePolicy::kAtEnd) {
ClipStack::ClipState oldState = cs.clipState();
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
cs.restore();
REPORTER_ASSERT(reporter, cs.clipState() == oldState,
"%s, restoring an empty save record should not change clip state: "
"expected %d but got %d",
name.c_str(), (int) oldState, (int) cs.clipState());
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
} else if (policy != SavePolicy::kNever) {
int restoreCount = policy == SavePolicy::kAtStart ? 1 : (int) order.size();
for (int i = 0; i < restoreCount; ++i) {
cs.restore();
}
// Should be wide open if everything is restored to base state
REPORTER_ASSERT(reporter, cs.clipState() == ClipStack::ClipState::kWideOpen,
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
"%s, restore should make stack become wide-open, not %d",
name.c_str(), (int) cs.clipState());
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
}
}
// All clip operations are commutative so applying actual elements in every possible order should
// always produce the same set of expected elements.
static void run_test_case(skiatest::Reporter* r, const TestCase& test) {
int n = (int) test.initialElements().size();
std::vector<int> order(n);
std::vector<int> stack(n);
// Initial order sequence and zeroed stack
for (int i = 0; i < n; ++i) {
order[i] = i;
stack[i] = 0;
}
auto runTest = [&]() {
static const SavePolicy kPolicies[] = { SavePolicy::kNever, SavePolicy::kAtStart,
SavePolicy::kAtEnd, SavePolicy::kBetweenEveryOp };
for (auto policy : kPolicies) {
test.run(order, policy, r);
}
};
// Heap's algorithm (non-recursive) to generate every permutation over the test case's elements
// https://en.wikipedia.org/wiki/Heap%27s_algorithm
runTest();
static constexpr int kMaxRuns = 720; // Don't run more than 6! configurations, even if n > 6
int testRuns = 1;
int i = 0;
while (i < n && testRuns < kMaxRuns) {
if (stack[i] < i) {
using std::swap;
if (i % 2 == 0) {
swap(order[0], order[i]);
} else {
swap(order[stack[i]], order[i]);
}
runTest();
stack[i]++;
i = 0;
testRuns++;
} else {
stack[i] = 0;
++i;
}
}
}
static SkPath make_octagon(const SkRect& r, SkScalar lr, SkScalar tb) {
SkPath p;
p.moveTo(r.fLeft + lr, r.fTop);
p.lineTo(r.fRight - lr, r.fTop);
p.lineTo(r.fRight, r.fTop + tb);
p.lineTo(r.fRight, r.fBottom - tb);
p.lineTo(r.fRight - lr, r.fBottom);
p.lineTo(r.fLeft + lr, r.fBottom);
p.lineTo(r.fLeft, r.fBottom - tb);
p.lineTo(r.fLeft, r.fTop + tb);
p.close();
return p;
}
static SkPath make_octagon(const SkRect& r) {
SkScalar lr = 0.3f * r.width();
SkScalar tb = 0.3f * r.height();
return make_octagon(r, lr, tb);
}
static constexpr SkIRect kDeviceBounds = {0, 0, 100, 100};
class NoOp : public GrDrawOp {
public:
static NoOp* Get() {
static NoOp gNoOp;
return &gNoOp;
}
private:
DEFINE_OP_CLASS_ID
NoOp() : GrDrawOp(ClassID()) {}
const char* name() const override { return "NoOp"; }
GrProcessorSet::Analysis finalize(const GrCaps&, const GrAppliedClip*, GrClampType) override {
return GrProcessorSet::EmptySetAnalysis();
}
void onPrePrepare(GrRecordingContext*, const GrSurfaceProxyView&, GrAppliedClip*, const
GrDstProxyView&, GrXferBarrierFlags, GrLoadOp) override {}
void onPrepare(GrOpFlushState*) override {}
void onExecute(GrOpFlushState*, const SkRect&) override {}
};
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
} // anonymous namespace
///////////////////////////////////////////////////////////////////////////////
// These tests use the TestCase infrastructure to define clip stacks and
// associated expectations.
// Tests that the initialized state of the clip stack is wide-open
DEF_TEST(ClipStack_InitialState, r) {
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
run_test_case(r, TestCase::Build("initial-state", SkIRect::MakeWH(100, 100)).finishTest());
}
// Tests that intersection of rects combine to a single element when they have the same AA type,
// or are pixel-aligned.
DEF_TEST(ClipStack_RectRectAACombine, r) {
using ClipState = skgpu::v1::ClipStack::ClipState;
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
SkRect pixelAligned = {0, 0, 10, 10};
SkRect fracRect1 = pixelAligned.makeOffset(5.3f, 3.7f);
SkRect fracRect2 = {fracRect1.fLeft + 0.75f * fracRect1.width(),
fracRect1.fTop + 0.75f * fracRect1.height(),
fracRect1.fRight, fracRect1.fBottom};
SkRect fracIntersect;
SkAssertResult(fracIntersect.intersect(fracRect1, fracRect2));
SkRect alignedIntersect;
SkAssertResult(alignedIntersect.intersect(pixelAligned, fracRect1));
// Both AA combine to one element
run_test_case(r, TestCase::Build("aa", kDeviceBounds)
.actual().aa().intersect()
.rect(fracRect1).rect(fracRect2)
.finishElements()
.expect().aa().intersect().rect(fracIntersect).finishElements()
.state(ClipState::kDeviceRect)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
// Both non-AA combine to one element
run_test_case(r, TestCase::Build("nonaa", kDeviceBounds)
.actual().nonAA().intersect()
.rect(fracRect1).rect(fracRect2)
.finishElements()
.expect().nonAA().intersect().rect(fracIntersect).finishElements()
.state(ClipState::kDeviceRect)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
// Pixel-aligned AA and non-AA combine
run_test_case(r, TestCase::Build("aligned-aa+nonaa", kDeviceBounds)
.actual().intersect()
.aa().rect(pixelAligned).nonAA().rect(fracRect1)
.finishElements()
.expect().nonAA().intersect().rect(alignedIntersect).finishElements()
.state(ClipState::kDeviceRect)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
// AA and pixel-aligned non-AA combine
run_test_case(r, TestCase::Build("aa+aligned-nonaa", kDeviceBounds)
.actual().intersect()
.aa().rect(fracRect1).nonAA().rect(pixelAligned)
.finishElements()
.expect().aa().intersect().rect(alignedIntersect).finishElements()
.state(ClipState::kDeviceRect)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
// Other mixed AA modes do not combine
run_test_case(r, TestCase::Build("aa+nonaa", kDeviceBounds)
.actual().intersect()
.aa().rect(fracRect1).nonAA().rect(fracRect2)
.finishElements()
.expectActual()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
}
// Tests that an intersection and a difference op do not combine, even if they would have if both
// were intersection ops.
DEF_TEST(ClipStack_DifferenceNoCombine, r) {
using ClipState = skgpu::v1::ClipStack::ClipState;
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
SkRect r1 = {15.f, 14.f, 23.22f, 58.2f};
SkRect r2 = r1.makeOffset(5.f, 8.f);
SkASSERT(r1.intersects(r2));
run_test_case(r, TestCase::Build("no-combine", kDeviceBounds)
.actual().aa().intersect().rect(r1)
.difference().rect(r2)
.finishElements()
.expectActual()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
}
// Tests that intersection of rects in the same coordinate space can still be combined, but do not
// when the spaces differ.
DEF_TEST(ClipStack_RectRectNonAxisAligned, r) {
using ClipState = skgpu::v1::ClipStack::ClipState;
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
SkRect pixelAligned = {0, 0, 10, 10};
SkRect fracRect1 = pixelAligned.makeOffset(5.3f, 3.7f);
SkRect fracRect2 = {fracRect1.fLeft + 0.75f * fracRect1.width(),
fracRect1.fTop + 0.75f * fracRect1.height(),
fracRect1.fRight, fracRect1.fBottom};
SkRect fracIntersect;
SkAssertResult(fracIntersect.intersect(fracRect1, fracRect2));
SkMatrix lm = SkMatrix::RotateDeg(45.f);
// Both AA combine
run_test_case(r, TestCase::Build("aa", kDeviceBounds)
.actual().aa().intersect().localToDevice(lm)
.rect(fracRect1).rect(fracRect2)
.finishElements()
.expect().aa().intersect().localToDevice(lm)
.rect(fracIntersect).finishElements()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
// Both non-AA combine
run_test_case(r, TestCase::Build("nonaa", kDeviceBounds)
.actual().nonAA().intersect().localToDevice(lm)
.rect(fracRect1).rect(fracRect2)
.finishElements()
.expect().nonAA().intersect().localToDevice(lm)
.rect(fracIntersect).finishElements()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
// Integer-aligned coordinates under a local matrix with mixed AA don't combine, though
run_test_case(r, TestCase::Build("local-aa", kDeviceBounds)
.actual().intersect().localToDevice(lm)
.aa().rect(pixelAligned).nonAA().rect(fracRect1)
.finishElements()
.expectActual()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
}
// Tests that intersection of two round rects can simplify to a single round rect when they have
// the same AA type.
DEF_TEST(ClipStack_RRectRRectAACombine, r) {
using ClipState = skgpu::v1::ClipStack::ClipState;
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
SkRRect r1 = SkRRect::MakeRectXY(SkRect::MakeWH(12, 12), 2.f, 2.f);
SkRRect r2 = r1.makeOffset(6.f, 6.f);
SkRRect intersect = SkRRectPriv::ConservativeIntersect(r1, r2);
SkASSERT(!intersect.isEmpty());
// Both AA combine
run_test_case(r, TestCase::Build("aa", kDeviceBounds)
.actual().aa().intersect()
.rrect(r1).rrect(r2)
.finishElements()
.expect().aa().intersect().rrect(intersect).finishElements()
.state(ClipState::kDeviceRRect)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
// Both non-AA combine
run_test_case(r, TestCase::Build("nonaa", kDeviceBounds)
.actual().nonAA().intersect()
.rrect(r1).rrect(r2)
.finishElements()
.expect().nonAA().intersect().rrect(intersect).finishElements()
.state(ClipState::kDeviceRRect)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
// Mixed do not combine
run_test_case(r, TestCase::Build("aa+nonaa", kDeviceBounds)
.actual().intersect()
.aa().rrect(r1).nonAA().rrect(r2)
.finishElements()
.expectActual()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
// Same AA state can combine in the same local coordinate space
SkMatrix lm = SkMatrix::RotateDeg(45.f);
run_test_case(r, TestCase::Build("local-aa", kDeviceBounds)
.actual().aa().intersect().localToDevice(lm)
.rrect(r1).rrect(r2)
.finishElements()
.expect().aa().intersect().localToDevice(lm)
.rrect(intersect).finishElements()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
run_test_case(r, TestCase::Build("local-nonaa", kDeviceBounds)
.actual().nonAA().intersect().localToDevice(lm)
.rrect(r1).rrect(r2)
.finishElements()
.expect().nonAA().intersect().localToDevice(lm)
.rrect(intersect).finishElements()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
}
// Tests that intersection of a round rect and rect can simplify to a new round rect or even a rect.
DEF_TEST(ClipStack_RectRRectCombine, r) {
using ClipState = skgpu::v1::ClipStack::ClipState;
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
SkRRect rrect = SkRRect::MakeRectXY({0, 0, 10, 10}, 2.f, 2.f);
SkRect cutTop = {-10, -10, 10, 4};
SkRect cutMid = {-10, 3, 10, 7};
// Rect + RRect becomes a round rect with some square corners
SkVector cutCorners[4] = {{2.f, 2.f}, {2.f, 2.f}, {0, 0}, {0, 0}};
SkRRect cutRRect;
cutRRect.setRectRadii({0, 0, 10, 4}, cutCorners);
run_test_case(r, TestCase::Build("still-rrect", kDeviceBounds)
.actual().intersect().aa().rrect(rrect).rect(cutTop).finishElements()
.expect().intersect().aa().rrect(cutRRect).finishElements()
.state(ClipState::kDeviceRRect)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
// Rect + RRect becomes a rect
SkRect cutRect = {0, 3, 10, 7};
run_test_case(r, TestCase::Build("to-rect", kDeviceBounds)
.actual().intersect().aa().rrect(rrect).rect(cutMid).finishElements()
.expect().intersect().aa().rect(cutRect).finishElements()
.state(ClipState::kDeviceRect)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
// But they can only combine when the intersecting shape is representable as a [r]rect.
cutRect = {0, 0, 1.5f, 5.f};
run_test_case(r, TestCase::Build("no-combine", kDeviceBounds)
.actual().intersect().aa().rrect(rrect).rect(cutRect).finishElements()
.expectActual()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
}
// Tests that a rect shape is actually pre-clipped to the device bounds
DEF_TEST(ClipStack_RectDeviceClip, r) {
using ClipState = skgpu::v1::ClipStack::ClipState;
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
SkRect crossesDeviceEdge = {20.f, kDeviceBounds.fTop - 13.2f,
kDeviceBounds.fRight + 15.5f, 30.f};
SkRect insideDevice = {20.f, kDeviceBounds.fTop, kDeviceBounds.fRight, 30.f};
run_test_case(r, TestCase::Build("device-aa-rect", kDeviceBounds)
.actual().intersect().aa().rect(crossesDeviceEdge).finishElements()
.expect().intersect().aa().rect(insideDevice).finishElements()
.state(ClipState::kDeviceRect)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
run_test_case(r, TestCase::Build("device-nonaa-rect", kDeviceBounds)
.actual().intersect().nonAA().rect(crossesDeviceEdge).finishElements()
.expect().intersect().nonAA().rect(insideDevice).finishElements()
.state(ClipState::kDeviceRect)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
}
// Tests that other shapes' bounds are contained by the device bounds, even if their shape is not.
DEF_TEST(ClipStack_ShapeDeviceBoundsClip, r) {
using ClipState = skgpu::v1::ClipStack::ClipState;
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
SkRect crossesDeviceEdge = {20.f, kDeviceBounds.fTop - 13.2f,
kDeviceBounds.fRight + 15.5f, 30.f};
// RRect
run_test_case(r, TestCase::Build("device-rrect", kDeviceBounds)
.actual().intersect().aa()
.rrect(SkRRect::MakeRectXY(crossesDeviceEdge, 4.f, 4.f))
.finishElements()
.expectActual()
.state(ClipState::kDeviceRRect)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
// Path
run_test_case(r, TestCase::Build("device-path", kDeviceBounds)
.actual().intersect().aa()
.path(make_octagon(crossesDeviceEdge))
.finishElements()
.expectActual()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
}
// Tests that a simplifiable path turns into a simpler element type
DEF_TEST(ClipStack_PathSimplify, r) {
using ClipState = skgpu::v1::ClipStack::ClipState;
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
// Empty, point, and line paths -> empty
SkPath empty;
run_test_case(r, TestCase::Build("empty", kDeviceBounds)
.actual().path(empty).finishElements()
.state(ClipState::kEmpty)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
SkPath point;
point.moveTo({0.f, 0.f});
run_test_case(r, TestCase::Build("point", kDeviceBounds)
.actual().path(point).finishElements()
.state(ClipState::kEmpty)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
SkPath line;
line.moveTo({0.f, 0.f});
line.lineTo({10.f, 5.f});
run_test_case(r, TestCase::Build("line", kDeviceBounds)
.actual().path(line).finishElements()
.state(ClipState::kEmpty)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
// Rect path -> rect element
SkRect rect = {0.f, 2.f, 10.f, 15.4f};
SkPath rectPath;
rectPath.addRect(rect);
run_test_case(r, TestCase::Build("rect", kDeviceBounds)
.actual().path(rectPath).finishElements()
.expect().rect(rect).finishElements()
.state(ClipState::kDeviceRect)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
// Oval path -> rrect element
SkPath ovalPath;
ovalPath.addOval(rect);
run_test_case(r, TestCase::Build("oval", kDeviceBounds)
.actual().path(ovalPath).finishElements()
.expect().rrect(SkRRect::MakeOval(rect)).finishElements()
.state(ClipState::kDeviceRRect)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
// RRect path -> rrect element
SkRRect rrect = SkRRect::MakeRectXY(rect, 2.f, 2.f);
SkPath rrectPath;
rrectPath.addRRect(rrect);
run_test_case(r, TestCase::Build("rrect", kDeviceBounds)
.actual().path(rrectPath).finishElements()
.expect().rrect(rrect).finishElements()
.state(ClipState::kDeviceRRect)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
}
// Tests that repeated identical clip operations are idempotent
DEF_TEST(ClipStack_RepeatElement, r) {
using ClipState = skgpu::v1::ClipStack::ClipState;
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
// Same rect
SkRect rect = {5.3f, 62.f, 20.f, 85.f};
run_test_case(r, TestCase::Build("same-rects", kDeviceBounds)
.actual().rect(rect).rect(rect).rect(rect).finishElements()
.expect().rect(rect).finishElements()
.state(ClipState::kDeviceRect)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
SkMatrix lm;
lm.setRotate(30.f, rect.centerX(), rect.centerY());
run_test_case(r, TestCase::Build("same-local-rects", kDeviceBounds)
.actual().localToDevice(lm).rect(rect).rect(rect).rect(rect)
.finishElements()
.expect().localToDevice(lm).rect(rect).finishElements()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
// Same rrect
SkRRect rrect = SkRRect::MakeRectXY(rect, 5.f, 2.5f);
run_test_case(r, TestCase::Build("same-rrects", kDeviceBounds)
.actual().rrect(rrect).rrect(rrect).rrect(rrect).finishElements()
.expect().rrect(rrect).finishElements()
.state(ClipState::kDeviceRRect)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
run_test_case(r, TestCase::Build("same-local-rrects", kDeviceBounds)
.actual().localToDevice(lm).rrect(rrect).rrect(rrect).rrect(rrect)
.finishElements()
.expect().localToDevice(lm).rrect(rrect).finishElements()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
// Same convex path, by ==
run_test_case(r, TestCase::Build("same-convex", kDeviceBounds)
.actual().path(make_octagon(rect)).path(make_octagon(rect))
.finishElements()
.expect().path(make_octagon(rect)).finishElements()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
run_test_case(r, TestCase::Build("same-local-convex", kDeviceBounds)
.actual().localToDevice(lm)
.path(make_octagon(rect)).path(make_octagon(rect))
.finishElements()
.expect().localToDevice(lm).path(make_octagon(rect))
.finishElements()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
// Same complicated path by gen-id but not ==
SkPath path; // an hour glass
path.moveTo({0.f, 0.f});
path.lineTo({20.f, 20.f});
path.lineTo({0.f, 20.f});
path.lineTo({20.f, 0.f});
path.close();
run_test_case(r, TestCase::Build("same-path", kDeviceBounds)
.actual().path(path).path(path).path(path).finishElements()
.expect().path(path).finishElements()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
run_test_case(r, TestCase::Build("same-local-path", kDeviceBounds)
.actual().localToDevice(lm)
.path(path).path(path).path(path).finishElements()
.expect().localToDevice(lm).path(path)
.finishElements()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
}
// Tests that inverse-filled paths are canonicalized to a regular fill and a swapped clip op
DEF_TEST(ClipStack_InverseFilledPath, r) {
using ClipState = skgpu::v1::ClipStack::ClipState;
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
SkRect rect = {0.f, 0.f, 16.f, 17.f};
SkPath rectPath;
rectPath.addRect(rect);
SkPath inverseRectPath = rectPath;
inverseRectPath.toggleInverseFillType();
SkPath complexPath = make_octagon(rect);
SkPath inverseComplexPath = complexPath;
inverseComplexPath.toggleInverseFillType();
// Inverse filled rect + intersect -> diff rect
run_test_case(r, TestCase::Build("inverse-rect-intersect", kDeviceBounds)
.actual().aa().intersect().path(inverseRectPath).finishElements()
.expect().aa().difference().rect(rect).finishElements()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
// Inverse filled rect + difference -> int. rect
run_test_case(r, TestCase::Build("inverse-rect-difference", kDeviceBounds)
.actual().aa().difference().path(inverseRectPath).finishElements()
.expect().aa().intersect().rect(rect).finishElements()
.state(ClipState::kDeviceRect)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
// Inverse filled path + intersect -> diff path
run_test_case(r, TestCase::Build("inverse-path-intersect", kDeviceBounds)
.actual().aa().intersect().path(inverseComplexPath).finishElements()
.expect().aa().difference().path(complexPath).finishElements()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
// Inverse filled path + difference -> int. path
run_test_case(r, TestCase::Build("inverse-path-difference", kDeviceBounds)
.actual().aa().difference().path(inverseComplexPath).finishElements()
.expect().aa().intersect().path(complexPath).finishElements()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
}
// Tests that clip operations that are offscreen either make the clip empty or stay wide open
DEF_TEST(ClipStack_Offscreen, r) {
using ClipState = skgpu::v1::ClipStack::ClipState;
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
SkRect offscreenRect = {kDeviceBounds.fRight + 10.f, kDeviceBounds.fTop + 20.f,
kDeviceBounds.fRight + 40.f, kDeviceBounds.fTop + 60.f};
SkASSERT(!offscreenRect.intersects(SkRect::Make(kDeviceBounds)));
SkRRect offscreenRRect = SkRRect::MakeRectXY(offscreenRect, 5.f, 5.f);
SkPath offscreenPath = make_octagon(offscreenRect);
// Intersect -> empty
run_test_case(r, TestCase::Build("intersect-combo", kDeviceBounds)
.actual().aa().intersect()
.rect(offscreenRect)
.rrect(offscreenRRect)
.path(offscreenPath)
.finishElements()
.state(ClipState::kEmpty)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
run_test_case(r, TestCase::Build("intersect-rect", kDeviceBounds)
.actual().aa().intersect()
.rect(offscreenRect)
.finishElements()
.state(ClipState::kEmpty)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
run_test_case(r, TestCase::Build("intersect-rrect", kDeviceBounds)
.actual().aa().intersect()
.rrect(offscreenRRect)
.finishElements()
.state(ClipState::kEmpty)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
run_test_case(r, TestCase::Build("intersect-path", kDeviceBounds)
.actual().aa().intersect()
.path(offscreenPath)
.finishElements()
.state(ClipState::kEmpty)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
// Difference -> wide open
run_test_case(r, TestCase::Build("difference-combo", kDeviceBounds)
.actual().aa().difference()
.rect(offscreenRect)
.rrect(offscreenRRect)
.path(offscreenPath)
.finishElements()
.state(ClipState::kWideOpen)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
run_test_case(r, TestCase::Build("difference-rect", kDeviceBounds)
.actual().aa().difference()
.rect(offscreenRect)
.finishElements()
.state(ClipState::kWideOpen)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
run_test_case(r, TestCase::Build("difference-rrect", kDeviceBounds)
.actual().aa().difference()
.rrect(offscreenRRect)
.finishElements()
.state(ClipState::kWideOpen)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
run_test_case(r, TestCase::Build("difference-path", kDeviceBounds)
.actual().aa().difference()
.path(offscreenPath)
.finishElements()
.state(ClipState::kWideOpen)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
}
// Tests that an empty shape updates the clip state directly without needing an element
DEF_TEST(ClipStack_EmptyShape, r) {
using ClipState = skgpu::v1::ClipStack::ClipState;
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
// Intersect -> empty
run_test_case(r, TestCase::Build("empty-intersect", kDeviceBounds)
.actual().intersect().rect(SkRect::MakeEmpty()).finishElements()
.state(ClipState::kEmpty)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
// Difference -> no-op
run_test_case(r, TestCase::Build("empty-difference", kDeviceBounds)
.actual().difference().rect(SkRect::MakeEmpty()).finishElements()
.state(ClipState::kWideOpen)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
SkRRect rrect = SkRRect::MakeRectXY({4.f, 10.f, 16.f, 32.f}, 2.f, 2.f);
run_test_case(r, TestCase::Build("noop-difference", kDeviceBounds)
.actual().difference().rrect(rrect).rect(SkRect::MakeEmpty())
.finishElements()
.expect().difference().rrect(rrect).finishElements()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
}
// Tests that sufficiently large difference operations can shrink the conservative bounds
DEF_TEST(ClipStack_DifferenceBounds, r) {
using ClipState = skgpu::v1::ClipStack::ClipState;
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
SkRect rightSide = {50.f, -10.f, 2.f * kDeviceBounds.fRight, kDeviceBounds.fBottom + 10.f};
SkRect clipped = rightSide;
SkAssertResult(clipped.intersect(SkRect::Make(kDeviceBounds)));
run_test_case(r, TestCase::Build("difference-cut", kDeviceBounds)
.actual().nonAA().difference().rect(rightSide).finishElements()
.expect().nonAA().difference().rect(clipped).finishElements()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
}
// Tests that intersections can combine even if there's a difference operation in the middle
DEF_TEST(ClipStack_NoDifferenceInterference, r) {
using ClipState = skgpu::v1::ClipStack::ClipState;
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
SkRect intR1 = {0.f, 0.f, 30.f, 30.f};
SkRect intR2 = {15.f, 15.f, 45.f, 45.f};
SkRect intCombo = {15.f, 15.f, 30.f, 30.f};
SkRect diff = {20.f, 6.f, 50.f, 50.f};
run_test_case(r, TestCase::Build("cross-diff-combine", kDeviceBounds)
.actual().rect(intR1, GrAA::kYes, SkClipOp::kIntersect)
.rect(diff, GrAA::kYes, SkClipOp::kDifference)
.rect(intR2, GrAA::kYes, SkClipOp::kIntersect)
.finishElements()
.expect().rect(intCombo, GrAA::kYes, SkClipOp::kIntersect)
.rect(diff, GrAA::kYes, SkClipOp::kDifference)
.finishElements()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
}
// Tests that multiple path operations are all recorded, but not otherwise consolidated
DEF_TEST(ClipStack_MultiplePaths, r) {
using ClipState = skgpu::v1::ClipStack::ClipState;
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
// Chosen to be greater than the number of inline-allocated elements and save records of the
// ClipStack so that we test heap allocation as well.
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
static constexpr int kNumOps = 16;
auto b = TestCase::Build("many-paths-difference", kDeviceBounds);
SkRect d = {0.f, 0.f, 12.f, 12.f};
for (int i = 0; i < kNumOps; ++i) {
b.actual().path(make_octagon(d), GrAA::kNo, SkClipOp::kDifference);
d.offset(15.f, 0.f);
if (d.fRight > kDeviceBounds.fRight) {
d.fLeft = 0.f;
d.fRight = 12.f;
d.offset(0.f, 15.f);
}
}
run_test_case(r, b.expectActual()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
b = TestCase::Build("many-paths-intersect", kDeviceBounds);
d = {0.f, 0.f, 12.f, 12.f};
for (int i = 0; i < kNumOps; ++i) {
b.actual().path(make_octagon(d), GrAA::kYes, SkClipOp::kIntersect);
d.offset(0.01f, 0.01f);
}
run_test_case(r, b.expectActual()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
}
// Tests that a single rect is treated as kDeviceRect state when it's axis-aligned and intersect.
DEF_TEST(ClipStack_DeviceRect, r) {
using ClipState = skgpu::v1::ClipStack::ClipState;
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
// Axis-aligned + intersect -> kDeviceRect
SkRect rect = {0, 0, 20, 20};
run_test_case(r, TestCase::Build("device-rect", kDeviceBounds)
.actual().intersect().aa().rect(rect).finishElements()
.expectActual()
.state(ClipState::kDeviceRect)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
// Not axis-aligned -> kComplex
SkMatrix lm = SkMatrix::RotateDeg(15.f);
run_test_case(r, TestCase::Build("unaligned-rect", kDeviceBounds)
.actual().localToDevice(lm).intersect().aa().rect(rect)
.finishElements()
.expectActual()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
// Not intersect -> kComplex
run_test_case(r, TestCase::Build("diff-rect", kDeviceBounds)
.actual().difference().aa().rect(rect).finishElements()
.expectActual()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
}
// Tests that a single rrect is treated as kDeviceRRect state when it's axis-aligned and intersect.
DEF_TEST(ClipStack_DeviceRRect, r) {
using ClipState = skgpu::v1::ClipStack::ClipState;
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
// Axis-aligned + intersect -> kDeviceRRect
SkRect rect = {0, 0, 20, 20};
SkRRect rrect = SkRRect::MakeRectXY(rect, 5.f, 5.f);
run_test_case(r, TestCase::Build("device-rrect", kDeviceBounds)
.actual().intersect().aa().rrect(rrect).finishElements()
.expectActual()
.state(ClipState::kDeviceRRect)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
// Not axis-aligned -> kComplex
SkMatrix lm = SkMatrix::RotateDeg(15.f);
run_test_case(r, TestCase::Build("unaligned-rrect", kDeviceBounds)
.actual().localToDevice(lm).intersect().aa().rrect(rrect)
.finishElements()
.expectActual()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
// Not intersect -> kComplex
run_test_case(r, TestCase::Build("diff-rrect", kDeviceBounds)
.actual().difference().aa().rrect(rrect).finishElements()
.expectActual()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
}
// Tests that scale+translate matrices are pre-applied to rects and rrects, which also then allows
// elements with different scale+translate matrices to be consolidated as if they were in the same
// coordinate space.
DEF_TEST(ClipStack_ScaleTranslate, r) {
using ClipState = skgpu::v1::ClipStack::ClipState;
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
SkMatrix lm = SkMatrix::Scale(2.f, 4.f);
lm.postTranslate(15.5f, 14.3f);
SkASSERT(lm.preservesAxisAlignment() && lm.isScaleTranslate());
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
// Rect -> matrix is applied up front
SkRect rect = {0.f, 0.f, 10.f, 10.f};
run_test_case(r, TestCase::Build("st+rect", kDeviceBounds)
.actual().rect(rect, lm, GrAA::kYes, SkClipOp::kIntersect)
.finishElements()
.expect().rect(lm.mapRect(rect), GrAA::kYes, SkClipOp::kIntersect)
.finishElements()
.state(ClipState::kDeviceRect)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
// RRect -> matrix is applied up front
SkRRect localRRect = SkRRect::MakeRectXY(rect, 2.f, 2.f);
SkRRect deviceRRect;
SkAssertResult(localRRect.transform(lm, &deviceRRect));
run_test_case(r, TestCase::Build("st+rrect", kDeviceBounds)
.actual().rrect(localRRect, lm, GrAA::kYes, SkClipOp::kIntersect)
.finishElements()
.expect().rrect(deviceRRect, GrAA::kYes, SkClipOp::kIntersect)
.finishElements()
.state(ClipState::kDeviceRRect)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
// Path -> matrix is NOT applied
run_test_case(r, TestCase::Build("st+path", kDeviceBounds)
.actual().intersect().localToDevice(lm).path(make_octagon(rect))
.finishElements()
.expectActual()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
}
// Tests that rect-stays-rect matrices that are not scale+translate matrices are pre-applied.
DEF_TEST(ClipStack_PreserveAxisAlignment, r) {
using ClipState = skgpu::v1::ClipStack::ClipState;
SkMatrix lm = SkMatrix::RotateDeg(90.f);
lm.postTranslate(15.5f, 14.3f);
SkASSERT(lm.preservesAxisAlignment() && !lm.isScaleTranslate());
// Rect -> matrix is applied up front
SkRect rect = {0.f, 0.f, 10.f, 10.f};
run_test_case(r, TestCase::Build("r90+rect", kDeviceBounds)
.actual().rect(rect, lm, GrAA::kYes, SkClipOp::kIntersect)
.finishElements()
.expect().rect(lm.mapRect(rect), GrAA::kYes, SkClipOp::kIntersect)
.finishElements()
.state(ClipState::kDeviceRect)
.finishTest());
// RRect -> matrix is applied up front
SkRRect localRRect = SkRRect::MakeRectXY(rect, 2.f, 2.f);
SkRRect deviceRRect;
SkAssertResult(localRRect.transform(lm, &deviceRRect));
run_test_case(r, TestCase::Build("r90+rrect", kDeviceBounds)
.actual().rrect(localRRect, lm, GrAA::kYes, SkClipOp::kIntersect)
.finishElements()
.expect().rrect(deviceRRect, GrAA::kYes, SkClipOp::kIntersect)
.finishElements()
.state(ClipState::kDeviceRRect)
.finishTest());
// Path -> matrix is NOT applied
run_test_case(r, TestCase::Build("r90+path", kDeviceBounds)
.actual().intersect().localToDevice(lm).path(make_octagon(rect))
.finishElements()
.expectActual()
.state(ClipState::kComplex)
.finishTest());
}
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
// Tests that a convex path element can contain a rect or round rect, allowing the stack to be
// simplified
DEF_TEST(ClipStack_ConvexPathContains, r) {
using ClipState = skgpu::v1::ClipStack::ClipState;
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
SkRect rect = {15.f, 15.f, 30.f, 30.f};
SkRRect rrect = SkRRect::MakeRectXY(rect, 5.f, 5.f);
SkPath bigPath = make_octagon(rect.makeOutset(10.f, 10.f), 5.f, 5.f);
// Intersect -> path element isn't kept
run_test_case(r, TestCase::Build("convex+rect-intersect", kDeviceBounds)
.actual().aa().intersect().rect(rect).path(bigPath).finishElements()
.expect().aa().intersect().rect(rect).finishElements()
.state(ClipState::kDeviceRect)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
run_test_case(r, TestCase::Build("convex+rrect-intersect", kDeviceBounds)
.actual().aa().intersect().rrect(rrect).path(bigPath).finishElements()
.expect().aa().intersect().rrect(rrect).finishElements()
.state(ClipState::kDeviceRRect)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
// Difference -> path element is the only one left
run_test_case(r, TestCase::Build("convex+rect-difference", kDeviceBounds)
.actual().aa().difference().rect(rect).path(bigPath).finishElements()
.expect().aa().difference().path(bigPath).finishElements()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
run_test_case(r, TestCase::Build("convex+rrect-difference", kDeviceBounds)
.actual().aa().difference().rrect(rrect).path(bigPath)
.finishElements()
.expect().aa().difference().path(bigPath).finishElements()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
// Intersect small shape + difference big path -> empty
run_test_case(r, TestCase::Build("convex-diff+rect-int", kDeviceBounds)
.actual().aa().intersect().rect(rect)
.difference().path(bigPath).finishElements()
.state(ClipState::kEmpty)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
run_test_case(r, TestCase::Build("convex-diff+rrect-int", kDeviceBounds)
.actual().aa().intersect().rrect(rrect)
.difference().path(bigPath).finishElements()
.state(ClipState::kEmpty)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
// Diff small shape + intersect big path -> both
run_test_case(r, TestCase::Build("convex-int+rect-diff", kDeviceBounds)
.actual().aa().intersect().path(bigPath).difference().rect(rect)
.finishElements()
.expectActual()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
run_test_case(r, TestCase::Build("convex-int+rrect-diff", kDeviceBounds)
.actual().aa().intersect().path(bigPath).difference().rrect(rrect)
.finishElements()
.expectActual()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
}
// Tests that rects/rrects in different coordinate spaces can be consolidated when one is fully
// contained by the other.
DEF_TEST(ClipStack_NonAxisAlignedContains, r) {
using ClipState = skgpu::v1::ClipStack::ClipState;
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
SkMatrix lm1 = SkMatrix::RotateDeg(45.f);
SkRect bigR = {-20.f, -20.f, 20.f, 20.f};
SkRRect bigRR = SkRRect::MakeRectXY(bigR, 1.f, 1.f);
SkMatrix lm2 = SkMatrix::RotateDeg(-45.f);
SkRect smR = {-10.f, -10.f, 10.f, 10.f};
SkRRect smRR = SkRRect::MakeRectXY(smR, 1.f, 1.f);
// I+I should select the smaller 2nd shape (r2 or rr2)
run_test_case(r, TestCase::Build("rect-rect-ii", kDeviceBounds)
.actual().rect(bigR, lm1, GrAA::kYes, SkClipOp::kIntersect)
.rect(smR, lm2, GrAA::kYes, SkClipOp::kIntersect)
.finishElements()
.expect().rect(smR, lm2, GrAA::kYes, SkClipOp::kIntersect)
.finishElements()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
run_test_case(r, TestCase::Build("rrect-rrect-ii", kDeviceBounds)
.actual().rrect(bigRR, lm1, GrAA::kYes, SkClipOp::kIntersect)
.rrect(smRR, lm2, GrAA::kYes, SkClipOp::kIntersect)
.finishElements()
.expect().rrect(smRR, lm2, GrAA::kYes, SkClipOp::kIntersect)
.finishElements()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
run_test_case(r, TestCase::Build("rect-rrect-ii", kDeviceBounds)
.actual().rect(bigR, lm1, GrAA::kYes, SkClipOp::kIntersect)
.rrect(smRR, lm2, GrAA::kYes, SkClipOp::kIntersect)
.finishElements()
.expect().rrect(smRR, lm2, GrAA::kYes, SkClipOp::kIntersect)
.finishElements()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
run_test_case(r, TestCase::Build("rrect-rect-ii", kDeviceBounds)
.actual().rrect(bigRR, lm1, GrAA::kYes, SkClipOp::kIntersect)
.rect(smR, lm2, GrAA::kYes, SkClipOp::kIntersect)
.finishElements()
.expect().rect(smR, lm2, GrAA::kYes, SkClipOp::kIntersect)
.finishElements()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
// D+D should select the larger shape (r1 or rr1)
run_test_case(r, TestCase::Build("rect-rect-dd", kDeviceBounds)
.actual().rect(bigR, lm1, GrAA::kYes, SkClipOp::kDifference)
.rect(smR, lm2, GrAA::kYes, SkClipOp::kDifference)
.finishElements()
.expect().rect(bigR, lm1, GrAA::kYes, SkClipOp::kDifference)
.finishElements()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
run_test_case(r, TestCase::Build("rrect-rrect-dd", kDeviceBounds)
.actual().rrect(bigRR, lm1, GrAA::kYes, SkClipOp::kDifference)
.rrect(smRR, lm2, GrAA::kYes, SkClipOp::kDifference)
.finishElements()
.expect().rrect(bigRR, lm1, GrAA::kYes, SkClipOp::kDifference)
.finishElements()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
run_test_case(r, TestCase::Build("rect-rrect-dd", kDeviceBounds)
.actual().rect(bigR, lm1, GrAA::kYes, SkClipOp::kDifference)
.rrect(smRR, lm2, GrAA::kYes, SkClipOp::kDifference)
.finishElements()
.expect().rect(bigR, lm1, GrAA::kYes, SkClipOp::kDifference)
.finishElements()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
run_test_case(r, TestCase::Build("rrect-rect-dd", kDeviceBounds)
.actual().rrect(bigRR, lm1, GrAA::kYes, SkClipOp::kDifference)
.rect(smR, lm2, GrAA::kYes, SkClipOp::kDifference)
.finishElements()
.expect().rrect(bigRR, lm1, GrAA::kYes, SkClipOp::kDifference)
.finishElements()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
// D(1)+I(2) should result in empty
run_test_case(r, TestCase::Build("rectD-rectI", kDeviceBounds)
.actual().rect(bigR, lm1, GrAA::kYes, SkClipOp::kDifference)
.rect(smR, lm2, GrAA::kYes, SkClipOp::kIntersect)
.finishElements()
.state(ClipState::kEmpty)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
run_test_case(r, TestCase::Build("rrectD-rrectI", kDeviceBounds)
.actual().rrect(bigRR, lm1, GrAA::kYes, SkClipOp::kDifference)
.rrect(smRR, lm2, GrAA::kYes, SkClipOp::kIntersect)
.finishElements()
.state(ClipState::kEmpty)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
run_test_case(r, TestCase::Build("rectD-rrectI", kDeviceBounds)
.actual().rect(bigR, lm1, GrAA::kYes, SkClipOp::kDifference)
.rrect(smRR, lm2, GrAA::kYes, SkClipOp::kIntersect)
.finishElements()
.state(ClipState::kEmpty)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
run_test_case(r, TestCase::Build("rrectD-rectI", kDeviceBounds)
.actual().rrect(bigRR, lm1, GrAA::kYes, SkClipOp::kDifference)
.rect(smR, lm2, GrAA::kYes, SkClipOp::kIntersect)
.finishElements()
.state(ClipState::kEmpty)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
// I(1)+D(2) should result in both shapes
run_test_case(r, TestCase::Build("rectI+rectD", kDeviceBounds)
.actual().rect(bigR, lm1, GrAA::kYes, SkClipOp::kIntersect)
.rect(smR, lm2, GrAA::kYes, SkClipOp::kDifference)
.finishElements()
.expectActual()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
run_test_case(r, TestCase::Build("rrectI+rrectD", kDeviceBounds)
.actual().rrect(bigRR, lm1, GrAA::kYes, SkClipOp::kIntersect)
.rrect(smRR, lm2, GrAA::kYes, SkClipOp::kDifference)
.finishElements()
.expectActual()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
run_test_case(r, TestCase::Build("rrectI+rectD", kDeviceBounds)
.actual().rrect(bigRR, lm1, GrAA::kYes, SkClipOp::kIntersect)
.rect(smR, lm2, GrAA::kYes, SkClipOp::kDifference)
.finishElements()
.expectActual()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
run_test_case(r, TestCase::Build("rectI+rrectD", kDeviceBounds)
.actual().rect(bigR, lm1, GrAA::kYes, SkClipOp::kIntersect)
.rrect(smRR, lm2, GrAA::kYes, SkClipOp::kDifference)
.finishElements()
.expectActual()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
}
// Tests that shapes with mixed AA state that contain each other can still be consolidated,
// unless they are too close to the edge and non-AA snapping can't be predicted
DEF_TEST(ClipStack_MixedAAContains, r) {
using ClipState = skgpu::v1::ClipStack::ClipState;
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
SkMatrix lm1 = SkMatrix::RotateDeg(45.f);
SkRect r1 = {-20.f, -20.f, 20.f, 20.f};
SkMatrix lm2 = SkMatrix::RotateDeg(-45.f);
SkRect r2Safe = {-10.f, -10.f, 10.f, 10.f};
SkRect r2Unsafe = {-19.5f, -19.5f, 19.5f, 19.5f};
// Non-AA sufficiently inside AA element can discard the outer AA element
run_test_case(r, TestCase::Build("mixed-outeraa-combine", kDeviceBounds)
.actual().rect(r1, lm1, GrAA::kYes, SkClipOp::kIntersect)
.rect(r2Safe, lm2, GrAA::kNo, SkClipOp::kIntersect)
.finishElements()
.expect().rect(r2Safe, lm2, GrAA::kNo, SkClipOp::kIntersect)
.finishElements()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
// Vice versa
run_test_case(r, TestCase::Build("mixed-inneraa-combine", kDeviceBounds)
.actual().rect(r1, lm1, GrAA::kNo, SkClipOp::kIntersect)
.rect(r2Safe, lm2, GrAA::kYes, SkClipOp::kIntersect)
.finishElements()
.expect().rect(r2Safe, lm2, GrAA::kYes, SkClipOp::kIntersect)
.finishElements()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
// Non-AA too close to AA edges keeps both
run_test_case(r, TestCase::Build("mixed-outeraa-nocombine", kDeviceBounds)
.actual().rect(r1, lm1, GrAA::kYes, SkClipOp::kIntersect)
.rect(r2Unsafe, lm2, GrAA::kNo, SkClipOp::kIntersect)
.finishElements()
.expectActual()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
run_test_case(r, TestCase::Build("mixed-inneraa-nocombine", kDeviceBounds)
.actual().rect(r1, lm1, GrAA::kNo, SkClipOp::kIntersect)
.rect(r2Unsafe, lm2, GrAA::kYes, SkClipOp::kIntersect)
.finishElements()
.expectActual()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
}
// Tests that a shape that contains the device bounds updates the clip state directly
DEF_TEST(ClipStack_ShapeContainsDevice, r) {
using ClipState = skgpu::v1::ClipStack::ClipState;
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
SkRect rect = SkRect::Make(kDeviceBounds).makeOutset(10.f, 10.f);
SkRRect rrect = SkRRect::MakeRectXY(rect, 10.f, 10.f);
SkPath convex = make_octagon(rect, 10.f, 10.f);
// Intersect -> no-op
run_test_case(r, TestCase::Build("rect-intersect", kDeviceBounds)
.actual().intersect().rect(rect).finishElements()
.state(ClipState::kWideOpen)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
run_test_case(r, TestCase::Build("rrect-intersect", kDeviceBounds)
.actual().intersect().rrect(rrect).finishElements()
.state(ClipState::kWideOpen)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
run_test_case(r, TestCase::Build("convex-intersect", kDeviceBounds)
.actual().intersect().path(convex).finishElements()
.state(ClipState::kWideOpen)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
// Difference -> empty
run_test_case(r, TestCase::Build("rect-difference", kDeviceBounds)
.actual().difference().rect(rect).finishElements()
.state(ClipState::kEmpty)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
run_test_case(r, TestCase::Build("rrect-difference", kDeviceBounds)
.actual().difference().rrect(rrect).finishElements()
.state(ClipState::kEmpty)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
run_test_case(r, TestCase::Build("convex-difference", kDeviceBounds)
.actual().difference().path(convex).finishElements()
.state(ClipState::kEmpty)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
}
// Tests that shapes that do not overlap make for an empty clip (when intersecting), pick just the
// intersecting op (when mixed), or are all kept (when diff'ing).
DEF_TEST(ClipStack_DisjointShapes, r) {
using ClipState = skgpu::v1::ClipStack::ClipState;
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
SkRect rt = {10.f, 10.f, 20.f, 20.f};
SkRRect rr = SkRRect::MakeOval(rt.makeOffset({20.f, 0.f}));
SkPath p = make_octagon(rt.makeOffset({0.f, 20.f}));
// I+I
run_test_case(r, TestCase::Build("iii", kDeviceBounds)
.actual().aa().intersect().rect(rt).rrect(rr).path(p).finishElements()
.state(ClipState::kEmpty)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
// D+D
run_test_case(r, TestCase::Build("ddd", kDeviceBounds)
.actual().nonAA().difference().rect(rt).rrect(rr).path(p)
.finishElements()
.expectActual()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
// I+D from rect
run_test_case(r, TestCase::Build("idd", kDeviceBounds)
.actual().aa().intersect().rect(rt)
.nonAA().difference().rrect(rr).path(p)
.finishElements()
.expect().aa().intersect().rect(rt).finishElements()
.state(ClipState::kDeviceRect)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
// I+D from rrect
run_test_case(r, TestCase::Build("did", kDeviceBounds)
.actual().aa().intersect().rrect(rr)
.nonAA().difference().rect(rt).path(p)
.finishElements()
.expect().aa().intersect().rrect(rr).finishElements()
.state(ClipState::kDeviceRRect)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
// I+D from path
run_test_case(r, TestCase::Build("ddi", kDeviceBounds)
.actual().aa().intersect().path(p)
.nonAA().difference().rect(rt).rrect(rr)
.finishElements()
.expect().aa().intersect().path(p).finishElements()
.state(ClipState::kComplex)
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
.finishTest());
}
DEF_TEST(ClipStack_ComplexClip, reporter) {
using ClipStack = skgpu::v1::ClipStack;
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
static constexpr float kN = 10.f;
static constexpr float kR = kN / 3.f;
// 4 rectangles that overlap by kN x 2kN (horiz), 2kN x kN (vert), or kN x kN (diagonal)
static const SkRect kTL = {0.f, 0.f, 2.f * kN, 2.f * kN};
static const SkRect kTR = {kN, 0.f, 3.f * kN, 2.f * kN};
static const SkRect kBL = {0.f, kN, 2.f * kN, 3.f * kN};
static const SkRect kBR = {kN, kN, 3.f * kN, 3.f * kN};
enum ShapeType { kRect, kRRect, kConvex };
SkRect rects[] = { kTL, kTR, kBL, kBR };
for (ShapeType type : { kRect, kRRect, kConvex }) {
for (int opBits = 6; opBits < 16; ++opBits) {
SkString name;
name.appendf("complex-%d-%d", (int) type, opBits);
SkRect expectedRectIntersection = SkRect::Make(kDeviceBounds);
SkRRect expectedRRectIntersection = SkRRect::MakeRect(expectedRectIntersection);
auto b = TestCase::Build(name.c_str(), kDeviceBounds);
for (int i = 0; i < 4; ++i) {
SkClipOp op = (opBits & (1 << i)) ? SkClipOp::kIntersect : SkClipOp::kDifference;
switch(type) {
case kRect: {
SkRect r = rects[i];
if (op == SkClipOp::kDifference) {
// Shrink the rect for difference ops, otherwise in the rect testcase
// any difference op would remove the intersection of the other ops
// given how the rects are defined, and that's just not interesting.
r.inset(kR, kR);
}
b.actual().rect(r, GrAA::kYes, op);
if (op == SkClipOp::kIntersect) {
SkAssertResult(expectedRectIntersection.intersect(r));
} else {
b.expect().rect(r, GrAA::kYes, SkClipOp::kDifference);
}
break; }
case kRRect: {
SkRRect rrect = SkRRect::MakeRectXY(rects[i], kR, kR);
b.actual().rrect(rrect, GrAA::kYes, op);
if (op == SkClipOp::kIntersect) {
expectedRRectIntersection = SkRRectPriv::ConservativeIntersect(
expectedRRectIntersection, rrect);
SkASSERT(!expectedRRectIntersection.isEmpty());
} else {
b.expect().rrect(rrect, GrAA::kYes, SkClipOp::kDifference);
}
break; }
case kConvex:
b.actual().path(make_octagon(rects[i], kR, kR), GrAA::kYes, op);
// NOTE: We don't set any expectations here, since convex just calls
// expectActual() at the end.
break;
}
}
// The expectations differ depending on the shape type
ClipStack::ClipState state = ClipStack::ClipState::kComplex;
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
if (type == kConvex) {
// The simplest case is when the paths cannot be combined together, so we expect
// the actual elements to be unmodified (both intersect and difference).
b.expectActual();
} else if (opBits) {
// All intersection ops were pre-computed into expectedR[R]ectIntersection
// - difference ops already added in the for loop
if (type == kRect) {
SkASSERT(expectedRectIntersection != SkRect::Make(kDeviceBounds) &&
!expectedRectIntersection.isEmpty());
b.expect().rect(expectedRectIntersection, GrAA::kYes, SkClipOp::kIntersect);
if (opBits == 0xf) {
state = ClipStack::ClipState::kDeviceRect;
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
}
} else {
SkASSERT(expectedRRectIntersection !=
SkRRect::MakeRect(SkRect::Make(kDeviceBounds)) &&
!expectedRRectIntersection.isEmpty());
b.expect().rrect(expectedRRectIntersection, GrAA::kYes, SkClipOp::kIntersect);
if (opBits == 0xf) {
state = ClipStack::ClipState::kDeviceRRect;
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
}
}
}
run_test_case(reporter, b.state(state).finishTest());
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
}
}
}
// ///////////////////////////////////////////////////////////////////////////////
// // These tests do not use the TestCase infrastructure and manipulate a
// // ClipStack directly.
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
// Tests that replaceClip() works as expected across save/restores
DEF_TEST(ClipStack_ReplaceClip, r) {
using ClipStack = skgpu::v1::ClipStack;
ClipStack cs(kDeviceBounds, nullptr, false);
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
SkRRect rrect = SkRRect::MakeRectXY({15.f, 12.25f, 40.3f, 23.5f}, 4.f, 6.f);
cs.clipRRect(SkMatrix::I(), rrect, GrAA::kYes, SkClipOp::kIntersect);
SkIRect replace = {50, 25, 75, 40}; // Is disjoint from the rrect element
cs.save();
cs.replaceClip(replace);
REPORTER_ASSERT(r, cs.clipState() == ClipStack::ClipState::kDeviceRect,
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
"Clip did not become a device rect");
REPORTER_ASSERT(r, cs.getConservativeBounds() == replace, "Unexpected replaced clip bounds");
const ClipStack::Element& replaceElement = *cs.begin();
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
REPORTER_ASSERT(r, replaceElement.fShape.rect() == SkRect::Make(replace) &&
replaceElement.fAA == GrAA::kNo &&
replaceElement.fOp == SkClipOp::kIntersect &&
replaceElement.fLocalToDevice == SkMatrix::I(),
"Unexpected replace element state");
// Restore should undo the replaced clip and bring back the rrect
cs.restore();
REPORTER_ASSERT(r, cs.clipState() == ClipStack::ClipState::kDeviceRRect,
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
"Unexpected state after restore, not kDeviceRRect");
const ClipStack::Element& rrectElem = *cs.begin();
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
REPORTER_ASSERT(r, rrectElem.fShape.rrect() == rrect &&
rrectElem.fAA == GrAA::kYes &&
rrectElem.fOp == SkClipOp::kIntersect &&
rrectElem.fLocalToDevice == SkMatrix::I(),
"RRect element state not restored properly after replace clip undone");
}
// Try to overflow the number of allowed window rects (see skbug.com/10989)
DEF_TEST(ClipStack_DiffRects, r) {
using ClipStack = skgpu::v1::ClipStack;
using SurfaceDrawContext = skgpu::v1::SurfaceDrawContext;
GrMockOptions options;
options.fMaxWindowRectangles = 8;
SkMatrixProvider matrixProvider = SkMatrix::I();
sk_sp<GrDirectContext> context = GrDirectContext::MakeMock(&options);
std::unique_ptr<SurfaceDrawContext> sdc = SurfaceDrawContext::Make(
context.get(), GrColorType::kRGBA_8888, SkColorSpace::MakeSRGB(),
SkBackingFit::kExact, kDeviceBounds.size(), SkSurfaceProps());
ClipStack cs(kDeviceBounds, &matrixProvider, false);
cs.save();
for (int y = 0; y < 10; ++y) {
for (int x = 0; x < 10; ++x) {
cs.clipRect(SkMatrix::I(), SkRect::MakeXYWH(10*x+1, 10*y+1, 8, 8),
GrAA::kNo, SkClipOp::kDifference);
}
}
GrAppliedClip out(kDeviceBounds.size());
SkRect drawBounds = SkRect::Make(kDeviceBounds);
GrClip::Effect effect = cs.apply(context.get(), sdc.get(), NoOp::Get(), GrAAType::kCoverage,
&out, &drawBounds);
REPORTER_ASSERT(r, effect == GrClip::Effect::kClipped);
REPORTER_ASSERT(r, out.windowRectsState().numWindows() == 8);
cs.restore();
}
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
// Tests that when a stack is forced to always be AA, non-AA elements become AA
DEF_TEST(ClipStack_ForceAA, r) {
using ClipStack = skgpu::v1::ClipStack;
ClipStack cs(kDeviceBounds, nullptr, true);
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
// AA will remain AA
SkRect aaRect = {0.25f, 12.43f, 25.2f, 23.f};
cs.clipRect(SkMatrix::I(), aaRect, GrAA::kYes, SkClipOp::kIntersect);
// Non-AA will become AA
SkPath nonAAPath = make_octagon({2.f, 10.f, 16.f, 20.f});
cs.clipPath(SkMatrix::I(), nonAAPath, GrAA::kNo, SkClipOp::kIntersect);
// Non-AA rects remain non-AA so they can be applied as a scissor
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
SkRect nonAARect = {4.5f, 5.f, 17.25f, 18.23f};
cs.clipRect(SkMatrix::I(), nonAARect, GrAA::kNo, SkClipOp::kIntersect);
// The stack reports elements newest first, but the non-AA rect op was combined in place with
// the first aa rect, so we should see nonAAPath as AA, and then the intersection of rects.
auto elements = cs.begin();
const ClipStack::Element& nonAARectElement = *elements;
REPORTER_ASSERT(r, nonAARectElement.fShape.isRect(), "Expected rect element");
REPORTER_ASSERT(r, nonAARectElement.fAA == GrAA::kNo,
"Axis-aligned non-AA rect ignores forceAA");
REPORTER_ASSERT(r, nonAARectElement.fShape.rect() == nonAARect,
"Mixed AA rects should not combine");
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
++elements;
const ClipStack::Element& aaPathElement = *elements;
REPORTER_ASSERT(r, aaPathElement.fShape.isPath(), "Expected path element");
REPORTER_ASSERT(r, aaPathElement.fShape.path() == nonAAPath, "Wrong path element");
REPORTER_ASSERT(r, aaPathElement.fAA == GrAA::kYes, "Path element not promoted to AA");
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
++elements;
const ClipStack::Element& aaRectElement = *elements;
REPORTER_ASSERT(r, aaRectElement.fShape.isRect(), "Expected rect element");
REPORTER_ASSERT(r, aaRectElement.fShape.rect() == aaRect,
"Mixed AA rects should not combine");
REPORTER_ASSERT(r, aaRectElement.fAA == GrAA::kYes, "Rect element stays AA");
++elements;
REPORTER_ASSERT(r, !(elements != cs.end()), "Expected only three clip elements");
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
}
// Tests preApply works as expected for device rects, rrects, and reports clipped-out, etc. as
// expected.
DEF_TEST(ClipStack_PreApply, r) {
using ClipStack = skgpu::v1::ClipStack;
ClipStack cs(kDeviceBounds, nullptr, false);
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
// Offscreen is kClippedOut
GrClip::PreClipResult result = cs.preApply({-10.f, -10.f, -1.f, -1.f}, GrAA::kYes);
REPORTER_ASSERT(r, result.fEffect == GrClip::Effect::kClippedOut,
"Offscreen draw is kClippedOut");
// Intersecting screen with wide-open clip is kUnclipped
result = cs.preApply({-10.f, -10.f, 10.f, 10.f}, GrAA::kYes);
REPORTER_ASSERT(r, result.fEffect == GrClip::Effect::kUnclipped,
"Wide open screen intersection is still kUnclipped");
// Empty clip is clipped out
cs.save();
cs.clipRect(SkMatrix::I(), SkRect::MakeEmpty(), GrAA::kNo, SkClipOp::kIntersect);
result = cs.preApply({0.f, 0.f, 20.f, 20.f}, GrAA::kYes);
REPORTER_ASSERT(r, result.fEffect == GrClip::Effect::kClippedOut,
"Empty clip stack preApplies as kClippedOut");
cs.restore();
// Contained inside clip is kUnclipped (using rrect for the outer clip element since paths
// don't support an inner bounds and anything complex is otherwise skipped in preApply).
SkRect rect = {10.f, 10.f, 40.f, 40.f};
SkRRect bigRRect = SkRRect::MakeRectXY(rect.makeOutset(5.f, 5.f), 5.f, 5.f);
cs.save();
cs.clipRRect(SkMatrix::I(), bigRRect, GrAA::kYes, SkClipOp::kIntersect);
result = cs.preApply(rect, GrAA::kYes);
REPORTER_ASSERT(r, result.fEffect == GrClip::Effect::kUnclipped,
"Draw contained within clip is kUnclipped");
// Disjoint from clip (but still on screen) is kClippedOut
result = cs.preApply({50.f, 50.f, 60.f, 60.f}, GrAA::kYes);
REPORTER_ASSERT(r, result.fEffect == GrClip::Effect::kClippedOut,
"Draw not intersecting clip is kClippedOut");
cs.restore();
// Intersecting clip is kClipped for complex shape
cs.save();
SkPath path = make_octagon(rect.makeOutset(5.f, 5.f), 5.f, 5.f);
cs.clipPath(SkMatrix::I(), path, GrAA::kYes, SkClipOp::kIntersect);
result = cs.preApply(path.getBounds(), GrAA::kNo);
REPORTER_ASSERT(r, result.fEffect == GrClip::Effect::kClipped && !result.fIsRRect,
"Draw with complex clip is kClipped, but is not an rrect");
cs.restore();
// Intersecting clip is kDeviceRect for axis-aligned rect clip
cs.save();
cs.clipRect(SkMatrix::I(), rect, GrAA::kYes, SkClipOp::kIntersect);
result = cs.preApply(rect.makeOffset(2.f, 2.f), GrAA::kNo);
REPORTER_ASSERT(r, result.fEffect == GrClip::Effect::kClipped &&
result.fAA == GrAA::kYes &&
result.fIsRRect &&
result.fRRect == SkRRect::MakeRect(rect),
"kDeviceRect clip stack should be reported by preApply");
cs.restore();
// Intersecting clip is kDeviceRRect for axis-aligned rrect clip
cs.save();
SkRRect clipRRect = SkRRect::MakeRectXY(rect, 5.f, 5.f);
cs.clipRRect(SkMatrix::I(), clipRRect, GrAA::kYes, SkClipOp::kIntersect);
result = cs.preApply(rect.makeOffset(2.f, 2.f), GrAA::kNo);
REPORTER_ASSERT(r, result.fEffect == GrClip::Effect::kClipped &&
result.fAA == GrAA::kYes &&
result.fIsRRect &&
result.fRRect == clipRRect,
"kDeviceRRect clip stack should be reported by preApply");
cs.restore();
}
// Tests the clip shader entry point
DEF_TEST(ClipStack_Shader, r) {
using ClipStack = skgpu::v1::ClipStack;
using SurfaceDrawContext = skgpu::v1::SurfaceDrawContext;
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
sk_sp<SkShader> shader = SkShaders::Color({0.f, 0.f, 0.f, 0.5f}, nullptr);
SkMatrixProvider matrixProvider = SkMatrix::I();
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
sk_sp<GrDirectContext> context = GrDirectContext::MakeMock(nullptr);
std::unique_ptr<SurfaceDrawContext> sdc = SurfaceDrawContext::Make(
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
context.get(), GrColorType::kRGBA_8888, SkColorSpace::MakeSRGB(),
SkBackingFit::kExact, kDeviceBounds.size(), SkSurfaceProps());
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
ClipStack cs(kDeviceBounds, &matrixProvider, false);
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
cs.save();
cs.clipShader(shader);
REPORTER_ASSERT(r, cs.clipState() == ClipStack::ClipState::kComplex,
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
"A clip shader should be reported as a complex clip");
GrAppliedClip out(kDeviceBounds.size());
SkRect drawBounds = {10.f, 11.f, 16.f, 32.f};
GrClip::Effect effect = cs.apply(context.get(), sdc.get(), NoOp::Get(), GrAAType::kCoverage,
&out, &drawBounds);
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
REPORTER_ASSERT(r, effect == GrClip::Effect::kClipped,
"apply() should return kClipped for a clip shader");
REPORTER_ASSERT(r, out.hasCoverageFragmentProcessor(),
"apply() should have converted clip shader to a coverage FP");
GrAppliedClip out2(kDeviceBounds.size());
drawBounds = {-15.f, -10.f, -1.f, 10.f}; // offscreen
effect = cs.apply(context.get(), sdc.get(), NoOp::Get(), GrAAType::kCoverage, &out2,
&drawBounds);
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
REPORTER_ASSERT(r, effect == GrClip::Effect::kClippedOut,
"apply() should still discard offscreen draws with a clip shader");
cs.restore();
REPORTER_ASSERT(r, cs.clipState() == ClipStack::ClipState::kWideOpen,
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
"restore() should get rid of the clip shader");
// Adding a clip shader on top of a device rect clip should prevent preApply from reporting
// it as a device rect
cs.clipRect(SkMatrix::I(), {10, 15, 30, 30}, GrAA::kNo, SkClipOp::kIntersect);
SkASSERT(cs.clipState() == ClipStack::ClipState::kDeviceRect); // test precondition
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
cs.clipShader(shader);
GrClip::PreClipResult result = cs.preApply(SkRect::Make(kDeviceBounds), GrAA::kYes);
REPORTER_ASSERT(r, result.fEffect == GrClip::Effect::kClipped && !result.fIsRRect,
"A clip shader should not produce a device rect from preApply");
}
// Tests apply() under simple circumstances, that don't require actual rendering of masks, or
// atlases. This lets us define the test regularly instead of a GPU-only test.
// - This is not exhaustive and is challenging to unit test, so apply() is predominantly tested by
// the GMs instead.
DEF_TEST(ClipStack_SimpleApply, r) {
using ClipStack = skgpu::v1::ClipStack;
using SurfaceDrawContext = skgpu::v1::SurfaceDrawContext;
SkMatrixProvider matrixProvider = SkMatrix::I();
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
sk_sp<GrDirectContext> context = GrDirectContext::MakeMock(nullptr);
std::unique_ptr<SurfaceDrawContext> sdc = SurfaceDrawContext::Make(
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
context.get(), GrColorType::kRGBA_8888, SkColorSpace::MakeSRGB(),
SkBackingFit::kExact, kDeviceBounds.size(), SkSurfaceProps());
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
ClipStack cs(kDeviceBounds, &matrixProvider, false);
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
// Offscreen draw is kClippedOut
{
SkRect drawBounds = {-15.f, -15.f, -1.f, -1.f};
GrAppliedClip out(kDeviceBounds.size());
GrClip::Effect effect = cs.apply(context.get(), sdc.get(), NoOp::Get(), GrAAType::kCoverage,
&out, &drawBounds);
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
REPORTER_ASSERT(r, effect == GrClip::Effect::kClippedOut, "Offscreen draw is clipped out");
}
// Draw contained in clip is kUnclipped
{
SkRect drawBounds = {15.4f, 16.3f, 26.f, 32.f};
cs.save();
cs.clipPath(SkMatrix::I(), make_octagon(drawBounds.makeOutset(5.f, 5.f), 5.f, 5.f),
GrAA::kYes, SkClipOp::kIntersect);
GrAppliedClip out(kDeviceBounds.size());
GrClip::Effect effect = cs.apply(context.get(), sdc.get(), NoOp::Get(), GrAAType::kCoverage,
&out, &drawBounds);
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
REPORTER_ASSERT(r, effect == GrClip::Effect::kUnclipped, "Draw inside clip is unclipped");
cs.restore();
}
// Draw bounds are cropped to device space before checking contains
{
SkRect clipRect = {kDeviceBounds.fRight - 20.f, 10.f, kDeviceBounds.fRight, 20.f};
SkRect drawRect = clipRect.makeOffset(10.f, 0.f);
cs.save();
cs.clipRect(SkMatrix::I(), clipRect, GrAA::kNo, SkClipOp::kIntersect);
GrAppliedClip out(kDeviceBounds.size());
GrClip::Effect effect = cs.apply(context.get(), sdc.get(), NoOp::Get(), GrAAType::kCoverage,
&out, &drawRect);
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
REPORTER_ASSERT(r, SkRect::Make(kDeviceBounds).contains(drawRect),
"Draw rect should be clipped to device rect");
REPORTER_ASSERT(r, effect == GrClip::Effect::kUnclipped,
"After device clipping, this should be detected as contained within clip");
cs.restore();
}
// Non-AA device rect intersect is just a scissor
{
SkRect clipRect = {15.3f, 17.23f, 30.2f, 50.8f};
SkRect drawRect = clipRect.makeOutset(10.f, 10.f);
SkIRect expectedScissor = clipRect.round();
cs.save();
cs.clipRect(SkMatrix::I(), clipRect, GrAA::kNo, SkClipOp::kIntersect);
GrAppliedClip out(kDeviceBounds.size());
GrClip::Effect effect = cs.apply(context.get(), sdc.get(), NoOp::Get(), GrAAType::kCoverage,
&out, &drawRect);
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
REPORTER_ASSERT(r, effect == GrClip::Effect::kClipped, "Draw should be clipped by rect");
REPORTER_ASSERT(r, !out.hasCoverageFragmentProcessor(), "Clip should not use coverage FPs");
REPORTER_ASSERT(r, !out.hardClip().hasStencilClip(), "Clip should not need stencil");
REPORTER_ASSERT(r, !out.hardClip().windowRectsState().enabled(),
"Clip should not need window rects");
REPORTER_ASSERT(r, out.scissorState().enabled() &&
out.scissorState().rect() == expectedScissor,
"Clip has unexpected scissor rectangle");
cs.restore();
}
// Analytic coverage FPs
auto testHasCoverageFP = [&](SkRect drawBounds) {
GrAppliedClip out(kDeviceBounds.size());
GrClip::Effect effect = cs.apply(context.get(), sdc.get(), NoOp::Get(), GrAAType::kCoverage,
&out, &drawBounds);
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
REPORTER_ASSERT(r, effect == GrClip::Effect::kClipped, "Draw should be clipped");
REPORTER_ASSERT(r, out.scissorState().enabled(), "Coverage FPs should still set scissor");
REPORTER_ASSERT(r, out.hasCoverageFragmentProcessor(), "Clip should use coverage FP");
};
// Axis-aligned rect can be an analytic FP
{
cs.save();
cs.clipRect(SkMatrix::I(), {10.2f, 8.342f, 63.f, 23.3f}, GrAA::kYes,
SkClipOp::kDifference);
testHasCoverageFP({9.f, 10.f, 30.f, 18.f});
cs.restore();
}
// Axis-aligned round rect can be an analytic FP
{
SkRect rect = {4.f, 8.f, 20.f, 20.f};
cs.save();
cs.clipRRect(SkMatrix::I(), SkRRect::MakeRectXY(rect, 3.f, 3.f), GrAA::kYes,
SkClipOp::kIntersect);
testHasCoverageFP(rect.makeOffset(2.f, 2.f));
cs.restore();
}
// Transformed rect can be an analytic FP
{
SkRect rect = {14.f, 8.f, 30.f, 22.34f};
SkMatrix rot = SkMatrix::RotateDeg(34.f);
cs.save();
cs.clipRect(rot, rect, GrAA::kNo, SkClipOp::kIntersect);
testHasCoverageFP(rot.mapRect(rect));
cs.restore();
}
// Convex polygons can be an analytic FP
{
SkRect rect = {15.f, 15.f, 45.f, 45.f};
cs.save();
cs.clipPath(SkMatrix::I(), make_octagon(rect), GrAA::kYes, SkClipOp::kIntersect);
testHasCoverageFP(rect.makeOutset(2.f, 2.f));
cs.restore();
}
}
// Must disable tessellation in order to trigger SW mask generation when the clip stack is applied.
static void disable_tessellation_atlas(GrContextOptions* options) {
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
options->fGpuPathRenderers = GpuPathRenderers::kNone;
options->fAvoidStencilBuffers = true;
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
}
DEF_GPUTEST_FOR_CONTEXTS(ClipStack_SWMask,
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
sk_gpu_test::GrContextFactory::IsRenderingContext,
r, ctxInfo, disable_tessellation_atlas) {
using ClipStack = skgpu::v1::ClipStack;
using SurfaceDrawContext = skgpu::v1::SurfaceDrawContext;
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
GrDirectContext* context = ctxInfo.directContext();
std::unique_ptr<SurfaceDrawContext> sdc = SurfaceDrawContext::Make(
context, GrColorType::kRGBA_8888, nullptr, SkBackingFit::kExact, kDeviceBounds.size(),
SkSurfaceProps());
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
SkMatrixProvider matrixProvider = SkMatrix::I();
std::unique_ptr<ClipStack> cs(new ClipStack(kDeviceBounds, &matrixProvider, false));
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
auto addMaskRequiringClip = [&](SkScalar x, SkScalar y, SkScalar radius) {
SkPath path;
path.addCircle(x, y, radius);
path.addCircle(x + radius / 2.f, y + radius / 2.f, radius);
path.setFillType(SkPathFillType::kEvenOdd);
// Use AA so that clip application does not route through the stencil buffer
cs->clipPath(SkMatrix::I(), path, GrAA::kYes, SkClipOp::kIntersect);
};
auto drawRect = [&](SkRect drawBounds) {
GrPaint paint;
paint.setColor4f({1.f, 1.f, 1.f, 1.f});
sdc->drawRect(cs.get(), std::move(paint), GrAA::kYes, SkMatrix::I(), drawBounds);
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
};
auto generateMask = [&](SkRect drawBounds) {
skgpu::UniqueKey priorKey = cs->testingOnly_getLastSWMaskKey();
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
drawRect(drawBounds);
skgpu::UniqueKey newKey = cs->testingOnly_getLastSWMaskKey();
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
REPORTER_ASSERT(r, priorKey != newKey, "Did not generate a new SW mask key as expected");
return newKey;
};
auto verifyKeys = [&](const std::vector<skgpu::UniqueKey>& expectedKeys,
const std::vector<skgpu::UniqueKey>& releasedKeys) {
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
context->flush();
GrProxyProvider* proxyProvider = context->priv().proxyProvider();
#ifdef SK_DEBUG
// The proxy providers key count fluctuates based on proxy lifetime, but we want to
// verify the resource count, and that requires using key tags that are debug-only.
SkASSERT(expectedKeys.size() > 0 || releasedKeys.size() > 0);
const char* tag = expectedKeys.size() > 0 ? expectedKeys[0].tag() : releasedKeys[0].tag();
GrResourceCache* cache = context->priv().getResourceCache();
int numProxies = cache->countUniqueKeysWithTag(tag);
REPORTER_ASSERT(r, (int) expectedKeys.size() == numProxies,
"Unexpected proxy count, got %d, not %d",
numProxies, (int) expectedKeys.size());
#endif
for (const auto& key : expectedKeys) {
auto proxy = proxyProvider->findOrCreateProxyByUniqueKey(key);
REPORTER_ASSERT(r, SkToBool(proxy), "Unable to find resource for expected mask key");
}
for (const auto& key : releasedKeys) {
auto proxy = proxyProvider->findOrCreateProxyByUniqueKey(key);
REPORTER_ASSERT(r, !SkToBool(proxy), "SW mask not released as expected");
}
};
// Creates a mask for a complex clip
cs->save();
addMaskRequiringClip(5.f, 5.f, 20.f);
skgpu::UniqueKey keyADepth1 = generateMask({0.f, 0.f, 20.f, 20.f});
skgpu::UniqueKey keyBDepth1 = generateMask({10.f, 10.f, 30.f, 30.f});
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
verifyKeys({keyADepth1, keyBDepth1}, {});
// Creates a new mask for a new save record, but doesn't delete the old records
cs->save();
addMaskRequiringClip(6.f, 6.f, 15.f);
skgpu::UniqueKey keyADepth2 = generateMask({0.f, 0.f, 20.f, 20.f});
skgpu::UniqueKey keyBDepth2 = generateMask({10.f, 10.f, 30.f, 30.f});
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
verifyKeys({keyADepth1, keyBDepth1, keyADepth2, keyBDepth2}, {});
// Release after modifying the current record (even if we don't draw anything)
addMaskRequiringClip(4.f, 4.f, 15.f);
skgpu::UniqueKey keyCDepth2 = generateMask({4.f, 4.f, 16.f, 20.f});
New GrClipStack supporting only intersect/difference Overview doc: https://docs.google.com/document/d/1ddIk74A1rL5Kj5kGcnInOYKVAXs3J2IsSgU5BLit0Ng/edit?usp=sharing This is the new clip stack that will replace GrClipStackClip. The doc link in the CL description has a much more detailed overview of what the strategy of the new clip stack is, but at a very high level: 1. Add a temporary #define that lets SkGpuDevice switch between the old stack and the new stack. For the new GrClipStack, it extends SkBaseDevice directly and has to implement all of the device clipping virtuals. - If you look from patchset 5 and earlier, the define defaults to on so I can test it on the bots, etc. but the plan will be for it to default to off when this lands so it's only running on unit tests. Then in a follow up, I'll turn it on for our bots but keep it off in chrome and android. If everything looks good, chrome can then be turned on. There is a more extensive migration plan for android because of the expanding clip ops, but that is covered at the end of the overview doc. 2. GrClipStack manages save/restore logic of the stack and extends GrClip, so the cpp file also includes code to apply a GrAppliedClip. At the moment the apply strategy is as close to that in GrReducedClip and GrClipStackClip as I could make it. Down the road, I think we can explore other analytic coverage options and a clip atlas that replaces the unified SW mask. - Once GrClipStack is enabled everywhere, it means GrReducedClip and GrClipStackClip can be deleted, so I'm not too worried about sharing code between the two. A lot is already shared through the use of GrSWMaskHelper and GrStencilMaskHelper. - SkClipStack and SkClipStackDevice are still used by the PDF and SVG backends, so they aren't necessarily deletable. 3. The GrClipStack only handles intersect and difference ops. It represents all geometric clip operations as an element. The stack itself is controlled by the "save record", which tracks aggregate bounds, valid elements, and the non-geometric clip shader. - When a new save record is pushed on the stack, older elements are inactive. This means they cannot be modified, since they may need to be activated again when the current save is popped off the stack. However, they can still affect the clip during application. - When a new element is pushed on the stack, older elements may be invalidated. This means they don't need to be considered any more because they are redundant with the new clip shape (e.g. nested round rect clips only have to keep the innermost valid). Bug: skia:10205 Change-Id: I68ccfd414033aa9014b102efaee3ad50a806f793 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308283 Commit-Queue: Michael Ludwig <michaelludwig@google.com> Reviewed-by: Robert Phillips <robertphillips@google.com> Reviewed-by: Brian Salomon <bsalomon@google.com>
2020-09-15 18:51:52 +00:00
verifyKeys({keyADepth1, keyBDepth1, keyCDepth2}, {keyADepth2, keyBDepth2});
// Release after restoring an older record
cs->restore();
verifyKeys({keyADepth1, keyBDepth1}, {keyCDepth2});
// Drawing finds the old masks at depth 1 still w/o making new ones
drawRect({0.f, 0.f, 20.f, 20.f});
drawRect({10.f, 10.f, 30.f, 30.f});
verifyKeys({keyADepth1, keyBDepth1}, {});
// Drawing something contained within a previous mask also does not make a new one
drawRect({5.f, 5.f, 15.f, 15.f});
verifyKeys({keyADepth1, keyBDepth1}, {});
// Release on destruction
cs = nullptr;
verifyKeys({}, {keyADepth1, keyBDepth1});
}