Add benchmarking test project and add text drawing benchmark tests #1876
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| test.Run(); | ||
| auto end = std::chrono::high_resolution_clock::now(); | ||
| test.PostRun(); | ||
| std::chrono::duration<Seconds> duration = end - start; |
There was a problem hiding this comment.
curious why are we doing manual time measurements.
gtest can do this for us.
There was a problem hiding this comment.
| Seconds netRuntime = std::accumulate(results.cbegin(), results.cend(), 0.0); | ||
| Seconds sampleMean = netRuntime / runCount; | ||
| Seconds sampleStandardDeviation = std::numeric_limits<Seconds>::infinity(); | ||
| if (runCount > 1) { |
There was a problem hiding this comment.
have we looked into the built in support gtest provides us?
There was a problem hiding this comment.
gtest already provides utility to help you measure running times.
It's part of the UnitTest lib.
There was a problem hiding this comment.
The only sort of timing utility I see in the gtest code is elapsed_time in TestResult. IIRC that won't work once @bbowman's changes are merged in with #1849. If that did work, we would still run into the problem of trying to only time part of the code run. Right now this only measures the time it takes to run the code in the Run method, but from what I can see gtest measures the entire construction of the test as well so we wouldn't be able to ignore setup/teardown time. Is there some other timing utility in gtest that I'm just not seeing/misunderstanding?
ms-jihua
reviewed
Jan 31, 2017
build/build.sln
Outdated
| @@ -1,6 +1,6 @@ | |||
| Microsoft Visual Studio Solution File, Format Version 12.00 | |||
| # Visual Studio 14 | |||
| VisualStudioVersion = 14.0.25123.0 | |||
There was a problem hiding this comment.
seems dangerous, what was the reasoning behind this?
There was a problem hiding this comment.
Unknown, didn't notice that changed when I added the project. Good catch
rajsesh
previously requested changes
Jan 31, 2017
| test.Run(); | ||
| auto end = std::chrono::high_resolution_clock::now(); | ||
| test.PostRun(); | ||
| std::chrono::duration<Seconds> duration = end - start; |
There was a problem hiding this comment.
Should compute in micro seconds if not nano seconds.
| } | ||
| }; | ||
|
|
||
| class CTFrameDrawGroup : public CTFrameDrawBase { |
There was a problem hiding this comment.
I'd like to also add tests that time the "create frame, line, etc" to this mix.
| TEST_BENCHMARK_F(CoreText, CTFrameDrawSingle); | ||
| TEST_BENCHMARK_F(CoreText, CTFrameDrawGroup); | ||
|
|
||
| class CTFrameDrawHuge : public TextBenchmarkBase { |
aballway
changed the title
|
Rebased onto develop as this new project depended on freetype which no longer exists 😸 Also ping! |
ms-jihua
reviewed
Feb 1, 2017
| Microseconds sampleStandardDeviation = std::sqrt(q / (runCount - 1)); | ||
|
|
||
| LOG_INFO( | ||
| "\nNet Runtime: %f microseconds\nSample Mean Runtime: %f microseconds\nSample Standard Deviation: %f microseconds\n%u test " |
There was a problem hiding this comment.
nit: probably out of scope, but I wonder if we would benefit from having a option to output in a csv format... probably overengineering if it looks like we'll eventually change the framework though. #Closed
ms-jihua
reviewed
Feb 1, 2017
|
|
||
| TEST_BENCHMARK_F(CoreText, CTFrameDrawComplete); | ||
|
|
||
| class CTFrameDrawYuge : public TextBenchmarkBase { |
There was a problem hiding this comment.
oh god why did you take @rajsesh-msft seriously
ms-jihua
reviewed
Feb 1, 2017
| } | ||
| }; | ||
|
|
||
| class CTFrameDrawGroup : public CTFrameDrawBase { |
There was a problem hiding this comment.
DrawGroup tests:
- should probably abstract out the PreRun PushBeginDraw/PostRun PopEndDraw to something common
- I'm kind of uncomfortable with the n=1 runs for the group tests. I realize it's 100 frames, and that these tests take a while, but EndDraw only happens once, meaning that some small temporary hiccup on the machine we run this on could lead to us overinterpreting noise. n=5 maybe?
ms-jihua
reviewed
Feb 2, 2017
| </ItemDefinitionGroup> | ||
| <ItemGroup> | ||
| <ClCompile Include="$(StarboardBasePath)\tests\unittests\Framework\Framework.cpp" /> | ||
| <ClCompile Include="..\..\..\..\tests\UnitTests\Benchmark\EntryPoint.cpp" /> |
There was a problem hiding this comment.
nit: standardize on StarboardBasePath; below too
ms-jihua
reviewed
Feb 2, 2017
build/build.sln
Outdated
| @@ -1,6 +1,6 @@ | |||
| Microsoft Visual Studio Solution File, Format Version 12.00 | |||
| # Visual Studio 14 | |||
| VisualStudioVersion = 14.0.25123.0 | |||
There was a problem hiding this comment.
Visual studio needs to chill 🍦
ms-jihua
reviewed
Feb 2, 2017
| } | ||
| }; | ||
|
|
||
| class CTFrameDrawGroup : public CTFrameDrawBase { |
There was a problem hiding this comment.
DrawGroup:
- n = 2 is better than n = 1 but probably still too few? other people can object if they think i'm being silly
- I would still like these Pre/PostRun functions to be abstracted out among the DrawGroup tests
There was a problem hiding this comment.
Oh I misunderstood what you were getting at. I'm not sure that they could be abstracted out w/o running into diamond problem. I changed FrameDrawGroup to use 5, which is giving a pretty consistently low stddev. I want to keep the default value low, but if others think it's worth bumping up I'll change it.
|
optional: For semi-vain reasons, can you post results from after #1878 as well? I'd like to see what the perf impact was more broadly. |
ms-jihua
reviewed
Feb 2, 2017
| CFRelease(descriptor); | ||
| CFRelease(font); | ||
| } |
There was a problem hiding this comment.
nit: add a newline at the end
rajsesh
reviewed
Feb 2, 2017
| CGContextSetFontSize(m_context, 20); | ||
|
|
||
| auto ctFont = woc::MakeAutoCF<CTFontRef>(CTFontCreateWithName(CFSTR("Arial"), 20, nullptr)); | ||
| EXPECT_TRUE(CTFontGetGlyphsForCharacters(ctFont, sc_chars, m_glyphs.data(), std::extent<decltype(sc_chars)>::value)); |
There was a problem hiding this comment.
CTFontGetGlyphsForCharacters [](start = 20, length = 28)
make one more test that benchmarks this api?
rajsesh
reviewed
Feb 2, 2017
| } | ||
|
|
||
| int main(int argc, char** argv) { | ||
| #ifdef WIN32 |
There was a problem hiding this comment.
We dont need this to run in reference platform, it that will make things easy w.r.t future integration to TAEF.
|
|
DHowett-MSFT
reviewed
Feb 2, 2017
tests/Benchmark/EntryPoint.cpp
Outdated
| // Align output horizontally | ||
| std::string spaces(m_maxNameWidth - 5, ' '); | ||
| std::cout << "Test Name" << spaces << "|" | ||
| << "Total Runtime" |
There was a problem hiding this comment.
You should almost universally prefer string formatting column widths instead of using \t.
There was a problem hiding this comment.
DHowett-MSFT
reviewed
Feb 2, 2017
tests/Benchmark/BenchmarkTest.h
Outdated
| // Can't be shared internally because parameterized tests need a different inheritance than non | ||
| template <typename T> | ||
| static void __RunTest(T& test) { | ||
| using Microseconds = double; |
There was a problem hiding this comment.
Microseconds [](start = 10, length = 12)
why not just use std::duration?
There was a problem hiding this comment.
like: typedef Microseconds to that, instead of feeding microseconds to it as a template arg
In reply to: 99247345 [](ancestors = 99247345)
There was a problem hiding this comment.
duration doesn't have sqrt defined so I needed to convert out to do the calucations for stdDev, which had me saving double rather than Microseconds for my values which I didn't do because I didn't want to obscure that they represent microseconds
DHowett-MSFT
reviewed
Feb 2, 2017
tests/Benchmark/BenchmarkTest.h
Outdated
| ::testing::internal::__RunTest(m_test); \ | ||
| } | ||
|
|
||
| #define BENCHMARK_REGISTER_TEST_P(test_name, test_class, generator, ...) \ |
There was a problem hiding this comment.
BENCHMARK_REGISTER_TEST_P [](start = 8, length = 25)
I don't get why we've cloven to the name TEST
We could use macros very similar to gtest's if we simply standardize on calling them BENCHMARK.
BENCHMARK(a, b, c)
BENCHMARK_F(a, b, c)
INSTANTIATE_BENCHMARK_CASE_P(a, b, c)
There was a problem hiding this comment.
DHowett-MSFT
reviewed
Feb 2, 2017
|
|
||
| inline void Run() { | ||
| [m_array release]; | ||
| } |
|
Prerun seems to be conflated with setup. What are the semantics for failures in the prerun method? Does this work with test_p? #Pending |
DHowett-MSFT
reviewed
Feb 3, 2017
tests/Benchmark/EntryPoint.cpp
Outdated
| #include <algorithm> | ||
|
|
||
| #ifdef WIN32 | ||
| #include <wrl\wrappers\corewrappers.h> |
There was a problem hiding this comment.
please standardize on / for include paths.
There is a visual studio option for this (!)
|
How do you mean conflated? This isn't testing anything in the sense of Unit Tests, so the only kind of failure that we should see would be a fatal failure, and that would just cause that particular test to fail without registering results. Yes, by using the BENCHMARK_REGISTER_CASE_P macro. In reply to: 277131527 [](ancestors = 277131527) |
|
Ping! |
DHowett-MSFT
reviewed
Feb 3, 2017
| std::vector<Microseconds> results(runCount); | ||
| @autoreleasepool { | ||
| for (size_t i = 0; i < runCount; ++i) { | ||
| test.PreRun(); |
There was a problem hiding this comment.
nit: is it worthwhile to aggregate time spent in setup and teardown?
There was a problem hiding this comment.
That is done in the constructor/destructor of test, this is for things that need to be done once before/after each test run
DHowett-MSFT
reviewed
Feb 3, 2017
tests/Benchmark/EntryPoint.cpp
Outdated
| } | ||
| } | ||
|
|
||
| if (!created) { |
There was a problem hiding this comment.
created [](start = 9, length = 7)
std::shared_ptr has an operator bool that you could use to check s_publisher's residence.
There was a problem hiding this comment.
DHowett-MSFT
reviewed
Feb 3, 2017
|
|
||
| void PostRun() { | ||
| _CGContextPopEndDraw(m_context); | ||
| } |
There was a problem hiding this comment.
If PostRun time doesn't count, you're ignoring how long it takes to actually draw here; is that your intent?
With the new CG model, every frame draw will be batched and flushed once down here.
There was a problem hiding this comment.
There was a problem hiding this comment.
:O no it is not holy moley
DHowett-MSFT
approved these changes
Feb 3, 2017
ms-jihua
approved these changes
Feb 3, 2017
|
@ms-jihua (before vs after drawing change) Yuge Savings HOly guacamole |
aballway
dismissed stale reviews from rajsesh and msft-Jeyaram
Raj gave an old fashioned ship it after the requested changes were addressed
rajsesh
pushed a commit
that referenced
this pull request
Feb 17, 2017
* D2DFactory Wrapper (#1085) * D2DFactory Wrapper * Fix path for wrapper includes in project file. * Moving starboard to mm file. * Address CR feedback, remove useless code, clean up D2D factories. * Moving header file to private location * Factory helper should only create single threaded D2D factories. * Actually fix the wrapper header location. * Add the CG/D2D design document. (#1112) * Implement an ID2D1RenderTarget-backed CGContext (#1126) There are wide swaths of code wrapped in #if 0 ... #endif blocks; they will be removed and repaired as part of the ongoing work on this branch. Issues have been logged for many TODOs, but the rest will be removed as part of the ongoing work on this branch. Currently broken: * path rendering * shadow * transparency layers * bitmap contexts that are not CALayer contexts * font rendering is left in CGContextCairo.mm and will be moved This merge is intended to unblock parallel work on CG/Direct2D. * Add CFCppBase, a helper class for implementing C++ CF types. * Switch CGContext/CGBitmapContext over to CppBase. * Initial implementation of CGPath Using ID2D1Geometries (#1138) * Initial implementation of CGPath Using ID2D1Geometries This work provides the foundation for creating a CGPath based on ID2D1PathGeometry and GeometrySink. A simple implementation for adding a simple line, moving the current location to a new point and retrieving some basic information from the Geometry sink has been added. Isolated unit tests have also been added to ensure these APIs are working as intended for these simple scenarios. * Addressing CR Feedback. * Reworking the __CGPath struct to use a nested implementation Also addressed some concerns with the state of geometry objects. * Fixing re-opening functionality by creating a new path and streaming contents of old path into it. * Addressing CR Feedback. * Cleaning up CGPath and removing unused code. * Refactor into using CFCppBase * Cleaning up a few things. * Addressing CR Feedback. Removing fully qualified names. Using inline helpers between D2D and CG. Fixing CGPathAddLines which will be tested more accurately in the future. Removing unnecessary dealloc method for CGPath base type. Fixing include->import. * Addming more helper functions and reverting bad test change. * Addressing CR Feedback. Fixing Inlines * Disabling tests that rely on CGPath in CoreText * CGContext DPI, parameter, stroking & interpolation improvements (#1181) * Make CGContext display scale-agnostic by propagating the scale into Direct2D's DPI. A CGContext rendering to a scaled surface will not apply a scale transformation. The scaling logic has been moved into render target creation. Fixes #1180. * Replace __CGContextDrawGeometry with RenderToCommandList+RenderCommandList. * Push CG Interpolation mode into Direct2D. References #1177. * Make all CG APIs nullptr-safe. * Move stroke style caching/computation into drawing state. * Fix CGContextConcatCTM (math was backwards.) * Perform major surgery on CGContext. * Fix improper use of internal functions in CGPath. (#1249) * Fix calculating exclusion zones based on CGPaths. * Implementation of CGImage using Windows Imaging Component. (#1185) - Implementation of CGImage via WIC - Remove circular dependencies between CGImage and UIImage - Implement UIImage interms of CGImage - Fix issues in UIImage - Increase code quality in UIImage - Remove libpng depencie of CGImage - Improvements to result.h Macros for conditional checks - Improve code quality for existing image interaction - Added unit tests * Add support for black and coloured shadows via ID2D1Effect. (#1239) * Add support for black and coloured shadows via ID2D1Effect. Closes #1237. * Apply the CTM transform for commandlist, device transform for final draw. Shadows project completely ignoring the context flip and any user translation. Therefore, we must translate all our drawing commands when we create them, but only apply the device scale in the final composition step. * Implement CGSizeApplyAffineTransform. * Allow a private consumer to control the shadow projection params. * Switch all internal interfaces to HR returns. * Implement the CGContext non-drawing path functions in terms of CGPath. (#1240) Curiously, CGContext path queries and CopyPath return untransformed units. Closes #1238. * Finish Simple line and status CGPath APIs. (#1255) * Finish Simple line and status CGPath APIs. Add additional tests to cover these scenarios. * Addressing CR Feedback. Releasing leaked paths in tests. * Addressing CR Feedback. Cleaning up nested conditionals and macro usage. * Addressing CR Feedback. * Removing unnecessary internal helper function. Adding extra test case for equality of paths with sub shapes. * Fixing access to internal members of implemtation for CGPath. * Rework __CGPath as per new encapsulation procedures. * Adding additional test case for possible sub shapes equality failure. * Implement drawing CGPaths through CGContext (#1307) * Implement drawing CGPaths through CGContext * Fix calls to return an HRESULT. * Address CR Feedback. * Fix poor comptr usage and leaks. Return to static initializer model. * Refine CGContext's user<->device coordinate space boundary (#1324) CGContext will now strive, wherever possible, to store its internal state in fully-device-transformed units. This transformation includes the device scale (flip+translate from CG to D2D) and the user scale. To enable this, this pull request also adds the ability to specify that coordinate system a drawn geometry is in: paths will use the premultiplied values, and user-space geometry values will use userspace->device transformation. This allows us to leverage the Direct2D effect stack without having to apply input and output transformations to the effects, and simplifies shadow rendering (we no longer transform for userspace in the commandlist and then device space when we draw it.) * Refactor CoreGraphics to use IWIC/D2D (#1332) - Full Implementation of CoreGraphics via IWIC/D2D - Refactor CoreGraphics #1299 #1227 #1246 #1292 #1345 * Update CGContext, CGPath and sorta CGImage for the CFCppBase changes. (#1356) * Disable CTFont.CreatePathForGlyph * Adapting the WIC factory creator to follow the outlined format in D2DWrapper (#1357) * Propagate context alpha into brushes for per-primitive draw, not layer * Add support for clipping (and EO clipping); add support for non-EO fill. (#1368) * Properly EO- or Winding-fill Geometries. * Add support for CGContext(EO)?Clip(). Fixes #1269. * Add CoreGraphics.Drawing UT and a bunch of image rendering tests. This change set introduces DRAW_TEST and DRAW_TEST_F, as well as a handful of image drawing test fixtures. Commands rendered to a test's context as part of its test body will be rasterized and saved in a file named TestImage.$TESTCASE.$TESTNAME.png The test driver's ideal interface is multi-modal; it needs to be able to generate a corpus of reference images on disk, and it needs to be able to diff them. For this, it needs a custom EntryPoint (included), and the ability to parse command-line arguments. The ideal default mode will be one that loads reference images from disk and fails tests if a number of pixels differ. That work is not yet complete. Hopefully, the draw tests here will be able to be plugged into another module that can perhaps render them to screen, or display them as part of a UI-driven flow for comparison and demoing purposes. Refs #1271. * Implementation of CGBitmapContext. (#1371) Added custom iwic bitmap to support copying, and backed data buffer. Added colour conversion format check for pixel formats. Added tests to draw, render and validate changes. CGImage changed to use OnLoad caching. * Changing IWIC image loading to on Load rather than on demand. (#1413) * Simplify geometry casting. (#1428) * Disable every drawing test manually instead of using exit(0). (#1434) * Cleaning up internal headers and moving CGImage implementation to source file (#1430) * Implementation of drawing an Image into context (#1414) Implement UIKit image drawing via WIC/D2D #1245 Added unit and semi functional tests for drawing into context and scaling. * Clean up internal headers; delete CGContextImpl. * Clean up CoreGraphics' public headers. This moves most of the types out of CoreGraphicsExport. * Automatically generate greenline images for tests and provide for test failure. (#1438) CoreGraphics.Drawing.UnitTests can now be invoked in one of two modes, mediated by command line flags: * `--compare[=/path/to/reference/folder]` (_default mode_) compare all rendered images against their baselines, failing the test if any images differ by even a single pixel. * `--generate [--out=/path/to/output/directory]` generate output images only (do not run comparisons.) A failing test will emit a brightly-colored diff image to `Greenlines.TestImage.CASE.TEST.png` in the _output directory_, and attach attributes to the gtest output XML data specifying where to find the images. A compatible test runner can load these images and display them side-by-side. **Open Questions/Concerns** * I do not like the design of `rgba/bgraPixels`. * I do not take stride into account; if the image is width-padded, the tests will ikely fail. * The reference platform loads images in RGBA, and we load them in BGRA. This is not an incompatibility, but it does make manual buffer manipulation dicey. * The fix here is to render the image into a NEW context, but that presents a few issues of its own: * We must trust our DrawImage function * We must further validate bitmap context, and trust our bitmap context as an integral part of test execution. * All tests will be disabled as of the merge of #1434. We can re-enable them and reintroduce reference images after build validation. Closes #1271. * remove export of internal function (#1458) * Fix internal and Objective-C use in CoreGraphics.DrawingTests. Fixed #1448. * Use CopyTo(p) instead of *p=Get(). Fixes #1453. * In CGImageDestination, never pair a Get with a Release. * *DO NOT* rely on the value of uninitialized memory. * zero out memory for backing buffer * Implement Arcs using D2D (#1384) * Implement arcs and curves in CGPath using D2D * Implement transformations in terms of D2D simplify. * Update curves, test app with Transformation options. * Fixing UI and other things. * Adding comments on arc math and updating cpputils * Use CGAffineTransform equality function instead of defining our own. * Fix const getter functions, reorder out pointer arguments * Fix failing coordinate tests due to stomped equality operator. * Re-Integration from develop->CGD2D - Removal of bridged IDisplayTexture #1337 - Getting it to build - Proper usage of the IDisplayTexture - Tests validated - Usage of proper colour format - Move image files to data folder and change tests to adapt. * [CGD2D] Add support for Transparency Layers. (#1496) This changeset introduces support for transparency layers. Transparency layers serve as render targets that are composed down to the backing surface in one fell swoop. To make way for this, CGContext was mildly refactored: * Existing private (denoted by `__CGContextX`) methods were moved into `__CGContext` the class as `__CGContext::X`. * Graphics state manipulation was promoted to `__CGContext`. * Some methods were promoted to drawing state methods (clipping, draw decision) * `__CGContextLayer`, a class that tracks a stack of drawing states and their destination buffers, was introduced. * CGContext's render target setup was moved into its constructor. * `globalAlpha` was reintroduced, as sometimes per-primitive alpha is just not enough. *Such as: layers are composed down all at once using the alpha set at the time the context was created.* Deficiencies: * Clipped layers are implemented in terms of a baseline clipping state applied to the layer. All primitives are drawn through a Direct2D **PushLayer** and **PopLayer** pair. * Alpha-composited layers are implemented in terms of the same. * There is some dodgy math witnessed on the reference platform that grows shadowed layers by some combination of shadow offset and width; an attempt was made to mimic this. * Minor layering violations that pass a `__CGContext` to a `CGContextXxx()` public function by way of **this**. Fixes #1427, #584. * Implement CGPathApply (#1507) * CGPathApply * Curve Apply Test Page * Update tests, address CR feedback, move geometry sink to individual file. * Addressing CR Feedback. * Fix no-op on cgpathapply when path is null. * Address CR Feedback * Single element point arrays are now just a variable * Ignoring clang warnings on ARM builds for WRL helpers. * Replace lost ID from last commit. * Remove bitmap context hack from NSTextContainer * Fix Affine Transform in CGPath (#1550) * Affine Transform * Remove excess com pointer usage from getters. * Update API status. * Enable Path Tests + Reference Images * Disabled AddEllipseInRect test until CGContextSetStrokeColor is supported. * Add support for copying subregions from CGIWICBitmap. References #1379. * Convert CGBitmapContext's image on output (if data isn't provided). * Convert ImageIO to use CGImage's WIC internals directly. Additionally, clean up some EXPECT/ASSERT/_MSG in ImageIO Tests. Fixes #1287. * Tidy CGImage's list of supported pixel and rendering formats. Add some more bitmap format tests, and adjust the rest of the tests to match. * Fix CoreGraphics.Drawing's image loader & premultiplied alpha. * The unpremultiplier here matches the algorithm used in WIC. * A test image has been added that will load unpremultiplied, but render premultiplied. * Add support+tests for clipping context rendering to an image mask. (#1528) This pull request adds support for clipping rendering into a context by an image or an image mask in terms of Direct2D opacity brushes. An image mask is a grayscale image whose pixel values range from 0 to 100%, and those values are taken as inverse alpha. That is: a fully black image (px=0.0) will be rendered through at alpha 1.0; a fully white pixel will be at alpha 0.0 and values between the two will be interpolated appropriately. Masks can be stacked. This is implemented in terms of a double composite operation; the two opacity brushes are combined and used to draw a black region into a new image. The new image will serve as the final bitmap backing the new opacity brush. Since Direct2D does not support b/w masks natively, an effort was undertaken to translate them into alpha channel masks. Black/white masks do, unfortunately, not work with CGBitmapContext as-is. We cannot render an image into a 1-component colorspace bitmap. This requires that this pull request be merged after #1529. Fixes #1421. Provides a base implementation for #1425. * Add support for CGImageCreateWithMask. (#1555) Fixes #1425. * Store CGContext's text position in its text matrix. (#1580) * Implementation of Tiled Images + tests (#1542) Implementation of Tiled Images + tests #1417 * Implementation of fill and stroke pattern brushes and tests #1422 * Implementation of set stroke and fill methods #1422 refactor existing code to use common functions. * Implementation of Linear gradients with full optional support (#1416) * Support opacity and interpolation for Drawing Images (#1557) * Add support for opacity and interpolation for drawing images #1556 * Reintegrate CoreGraphics and Accelerate.framework. (#1680) Fix some dodgy memory management in Accelerate. * Fix Accelerate.UnitTests. Fixes #1285. * Fix (and update) UIGraphicsFunctions. (#1681) Fixes #1642. * Do not use UIColor in CoreText! * Fix CGImage to not copy CGDataProvider. Fix CGDataProvider in vImage. * CoreImage:Implementation of CIImage using CGImage + test (#1625) * CoreImage:Implementation of CIImage using CGImage + test * Proper annotations for implemented methods (#1719) * Implementation of JPEG & PNG representation and tests * #1338 Adding support to copy images to/from clipboard * Add woc::AutoCF<T>. It's like AutoId<T> for CFTypes. Just like AutoId<T>, it takes a lifetime trait template: CFLifetimeRetain for CFRetain/CFRelease, and CFLifetimeUnsafe for straight assignment. It builds upon and replaces woc::unique_cf<T>. * It is copyable (and can share ownership) * AutoCF<T> can be passed anywhere T* is required. * It has a fully-loaded operator& so you can pass it wherever a T** is required. The old value will be released. * It has counterparts in StrongCF and UnsafeCF, just like StrongId and UnsafeId. By default, a StrongCF will *share ownership* of a CFType. To transfer ownership, construct it with std::move(cf) or use woc::MakeStrongCF<T>(CFCreateXxx(...)). As a migration measure, woc::unique_cf is still provided, though it is now based on StrongCF. It retains its "transfer ownership on construct" semantics and provides a few shim methods to adapt it to the AutoCF/AutoID world (release -> detach, reset -> attach). (cherry picked from commit a64f482) * Fix fallout from the merge of #1692. * Migrate the text drawing code from Cairo to CGD2D. * Fix the default fill/stroke color values from the reference platform. * #1733 CGContextClearRect: Support to clear rect if clipping is not set (#1743) * Fix our WIC<->CG color format table again. Fixes #1694. * Work around #1769 by doing the DPI calculation ourselves. * Fix Figure Creation in CGPath (#1738) * Fix Figure Creation in CGPath * Fix private class member naming. * Merge with CGD2D * Check for null geometry on figure end calls * Plug memory leak in __DWriteTextLayoutApplyFont (#1788) Fixes #1766 * Fixing ApiAnalyzer errors. (#1794) * NSTextContainer handle exclusion zones extending outside of area (#1792) Fixes #1779 * getNetworkConnectivityLevel can throw under some circumstances. Handle (#1795) * getNetworkConnectivityLevel can throw under some circumstances. Handle these and treat the connection status as 'not connected'. * Fix leak in NSRunLoop _statesForMode: helper (#1793) * Fix leak in NSRunLoop _statesForMode: helper * just use autorelease for helper; remove additional unused method from NSRunLoop (which would double release if called) * SC callbacks must by default happen on the main thread unless the app (#1801) specifies a different queue. This fixes a crash in reachability sample. * Fix drawing tests on OSX (#1815) Need to be explicit about stealing the deltaImage when doing a move of our ImageDelta Fixes #1805 * UX Functional test pattern - exercising sample app's UIViewController within a functional test (#1771) CHANGES: 1. Created a new XAMLCatalog UIViewController with 2 controls (slider, UITextField) to manipulate a UIButton's label 2. This UIViewController's exposes a public UIButton property that can be manipulated from a functional test (UIButtonTests) 3. Created UX helper class, ViewControllerTestHelper, to display visuals using pushVC/dismissVC 4. Created UX helper class, XamlEventSubscription, to register/unregister XAML property change events 5. Updated FunctionalTest VCXPROJ to move external VCs into its own folder TODO: 1. Migrate NSCondition to Win32 event mechanism * CoreText Performance: Call ID2D1RenderTarget::BeginDraw()/EndDraw() f… (#1705) * CoreText Performance: Call ID2D1RenderTarget::BeginDraw()/EndDraw() fewer times Fixes #1620 * - Added Begin/EndDraw pairs to a few more places - Mitigated an issue with cairo where ID2D1RenderTarget would sometimes cache a before state during begin draw, and wipe out any changes cairo made, breaking APIs like CGContextFillRect. - Changed implementation/naming of the 'escape the current begin/end draw stack' functions (cherry picked from commit d5a86b9) * Merge #1705 to CGD2D - Add PushBeginDraw/PopEndDraw pairs to UISegment, UIImage (not on develop since this likely would've actually hurt performance there) - Add Escape/Unescape pairs to CGContext areas where the target is changed Fixes #1635 cr feedback cr feedback (cherry picked from commit 0ac5491) * Remove Cairo and unused dependencies (#1797) Fixes #1455 * [CGD2D] Render buffer-shared G8 images as A8. (#1804) This is riddled with caveats, many of which are documented in the code. A first pass attempt was made to implement a format converter that would run per-primitive. It's expensive, but it did get the job done. Unfortunately, asking WIC to convert a PRGBA buffer to a G8 buffer causes it to discard all alpha data -- even if premultiplied, the luminance of partially-transparent pixels is not appropriately reduced in the final G8 image. This impacts alpha-heavy uses such as grayscale font rendering. The A8 implementation was chosen for maximal compatibiltiy with Cocos2D. Cocos2D renders white (!) text into a "grayscale" buffer that it then passes off to OpenGLES as a `GL_ALPHA8` buffer. Hah. The A8 "hack" also represents what we currently have on the develop branch. Fixes #1747. * Use custom font for stable version and copyright (#1799) Fixes #1798 * Change parameterized test to use custom font (thanks Dustin :D) * Initialize m_lastPoint; fixes intermittent test failure. * Porting ObjectiveC language services VSIX to VS2017 (#1745) * Make sure our touch/point conversion works in middleware (C#-hosted, etc.) scenarios. Controls that are placed within arbitrary Xaml scenes don't have a superview chain up to a valid UIWindow, so touch point conversion is failing. The fix is to - in middleware scenarios - use the app's main UIWindow for any UIView that doesn't have a superview. (#1821) Fixes #1818. * Fix various Uikit memory leaks (#1803) * Fixing some UIKit/WinRT projections memory leaks. Any time we allocate a WinRT projected object via make*, we are responsible for freeing it. * Fixing leak in Autolayout and its usage of ClSlackVariables. * Updating libcassowary build output with memory leak fix. * Incorporating CR feedback. * Allow _UIPopupViewController to present itself in middleware scenarios. (#1832) A recent fix/change to UIViewController (75effb9) regressed one of our middleware scenarios. UIViewController now makes sure that one is presenting over a currently-visible ViewController. However, in middleware scenarios, that won't hold true, so we must loosen the check in such cases. Fixes #1831. * Fix additional leaks in UIKit (#1789) * Fix additional leaks in UIKit * just add autorelease to paragraphStyle helper * address CR (prefer local StrongId to adding a bare release in UIDatePicker; add ARC enabled assertions). fix additional leaks in UITabBarController. * ObjC2Winmd: Support for async delegates with multiple parameters (#1833) * Fix a missed endianness flip in #1748. Fixes #1837. * Fix various leaks in Foundation (#1834) Thanks to @ehren. * Migrate foundation tests out of woccatalog (#1826) Moves Foundation Tests, which includes task_info, URL Session, and UserDefaults, and moves them to a new test app Foundation-Dev in tests/testapps Fixes #1577 * Migrate foundation tests out of woccatalog * Even more relative path * Updated GLKitComplex sample project with missing textures. (#1842) * Add data files to xcodeproject * Updated glkit project with missing textures. Fixes #1840. * Fixing Github issue 1602 and 1689 - sizeThatFits and adjustFontSizeToFitWidth does not work on UILabel * Revert "Redo locking and ref counting in NSOperationQueue to be much safer. A… (#1676)" This reverts commit 1db8c66. * Revert "Fix various leaks in Foundation (#1834)" This reverts commit 3a65494. * Fix #1827 by flipping the context *properly* in UIImage draw. * Reimplement CGContextDrawImage in terms of _CGContextDrawImageRect. This fixes the heavy drawing overhead on 9patch images, which used to apply a context-global clip and draw the image in full nine separate times. Now we get to use D2D's native image region subsetting support! * Created UX test helper class and simplified the tests using the new UX pattern (#1855) * Created UX test helper class and simplified the tests using the new UX pattern CHANGES: 1. Created UXEvent class, wrapping NSCondition, and UIColor and WUXMSolidBrush comparison method 2. Added more controls to manipulate a UIButton via XAMLCatalog's viewcontroller 3. Applied new UX test pattern to CALayerAppearance tests Fixes #1764, #1757 * Fix CGPathEqualToPath Implementation (#1813) * Fix CGPathEqualToPath * Address naming feedback and coding conventions. * Address CR Feedback. * Fix overuse of m_ and provide more proper names for parameters * Fix islandwood news crash in article view when panning (#1857) * Some improvements to RETURN_RESULT_* macros (#1862) * Some improvements to RETURN_RESULT_* macros * some nit picking * Use _CGContextPushBegin/PopEndDraw for Drawing Tests (#1870) * Use _CGContextPushBegin/PopEndDraw for Drawing Tests - Since we generally expect real use-cases to use these functions for perf purposes, we should also validate that our drawing is correct when using them. - Fixed a stray raw ->BeginDraw() ->EndDraw() in CGContext.mm that was missed and could cause crashes if used under PushBegin/PopEndDraw Fixes #1869 * better format a comment * Adding macros to enchance return and failure checks. (#1871) * NSURLSession should throw for creating tasks with nil request or resume data (#1860) Fixes #1844 * Check for nil before trying to resume * CR feedback * Fix starting position of arc in bezierPathWithArcCenter (#1864) * Fixing more UIKit memory leaks (#1863) * Fixing leak in UIKit.Xaml controls, and a leak in UIButton. * Adding UIButton memory leak tests. * Switching to fancy new event helper. * Add a slight delay after freeing the UIBUtton to prevent a very intermittent test failure (failed twice out of 500 times). * Switching to StrongId for VC. * Incorporating CR feedback. * Fix various leaks in Foundation (#1873) Fixes leaks in NSLog and internal logging funcs, NSString and NSMutableString leaks, and an NSTimer leak. * Remove some dead code from bff7cce. * If the user provides bitmap data, a bitmap context must copy it out. * [vImage] Don't hire a bitmap context to do our dirty work. Converting an arbitrary buffer of bytes into an image should be handled by CGImage. Using CGBitmapContext and its rather more limited set of formats is bound to fail more often. * Fix the Accelerate samples to use 32bpp images & proper byte ordering. * Fix a missed swap from the format table introduced in 2122f2a. * Winmd2ObjC update: Fixed inconsistencies with create method and added annotations. (#1883) * Enable vertical pixel-snapping when drawing glyph runs (#1890) - Devices with vertical antialiasing would render text as blurry if rendered on a fractional vertical coordinate. - CGContext functions that draw glyphs do not necessarily use coordinates from our custom text renderer, which already has pixel snapping enabled. - As such, manually round the y-translation in CGContext::DrawGlyphRun() - Updated reference images for two drawing tests dealing with line height. Fixes #1594 * Remove EbrCenterTextInRectVertically (#1882) Changes UITabBarButton to directly center text vertically in drawRect: Removes dead code from UITextView-sizeThatFits: Fixes #1812 * Change targets less in CGContext (#1878) * Change targets less in CGContext - CGContext previously changed targets for all draw primitive functions. Since changing targets causes an immediate flush, this significantly negatively impacted performance. Changed implementation to only change targets when needed (when shadows are enabled) - Added drawing tests for a few edge cases regarding shadows and clipping. - Removed unnecessary Escape/Unescape functions - these were not necessary to change targets as previously surmised. Fixes #1810 * - Enabled all shadow tests, using our current output (as opposed to reference platform) as reference images - Make the reversion of transform to the identity matrix safer - Misc cleanup * CR feedback * Rewrite portions of EbrFile handling to use UTF16 (#1891) * Rewrite path mapper to simplify code and support UTF16 path names. * Rewrite Ebr file apis to use new path mapper. * Remove dead code. * fixed unit tests to build after refactor. * Add NSFileManager test to use special characters. * EbrGet/SetWritablePath are now IwGet/SetWritablePath, and support UTF-16. Fixes #1875 * Updating functionaltest.ps1 so it works in VSTS. (#1896) * Disabling some NSDecimal subtraction tests on OSX (#1908) * Update NSURLSessionConfiguration defaults to support OSX 10.12 (#1905) #1905 * Objc2Winmd binary update: (#1913) * Objc2Winmd binary update: Fixed memory leak issues with IAsyncOperation implementation * ObjC2Winmd binary update: Removed unnecessary comments * Add benchmarking test project and add text drawing benchmark tests (#1876) Creates a new test project Benchmark in the Unit Test group, which leverages GTest to run automated benchmark tests. Simple tests can be done with the TEST_BENCHMARK macro, or more complex ones can be used by creating classes to do setup and teardown so only specific parts of code are measured. * Turning of clang modules. (#1914) * Moving windows specific archival test(BackwardsCompatibilityWithOldWinObjCArchives) to windows only unit tests (#1907) * This change implement UILabel minimumScale (#1918) * implement UILabel minimumScaleFactor * revert unncessary change to diagnose #1894 * CR comments * comments * SetTempFolder() should take wchar_t instead of char so that we don't (#1916) * SetTempFolder() should take wchar_t instead of char so that we don't lose encoding information. Fixes #1915 * Update usage of path names to UTF8 instead of CString * CR feedback * Update Implementation of NSURL.deletingLastPathExtension to OSX 10.12 (#1906) * Improve performance in __CF_IsBridgedObject/__CF_IsCFObject (#1923) * Improve performance in __CF_IsBridgedObject/__CF_IsCFObject - __CF_IsCFObject() calls __CF_IsBridgedObject() - Previously, __CF_IsBridgedObject() would do an O(n^2) comparison against the entire runtime class table. (O(n) traversal of table * O(n) check of inheritance tree) - These two functions are used in CF memory management functions, such as CFRetain/Release, and due to their prevalence, took up substantial portions of CPU time. - Changed bridged classes to adopt an empty 'bridged object' protocol when they are bridged/added to the class table. This protocol can be quickly checked against, and obviates the need to traverse the table. - Removed some redundant _cfisa checks in _CF_IsCFObject Fixes #1490 * - Remove use of objective-C * - Also handle NSCFType * Switch our D2D factory to be multithreaded. (#1911) Fixes #1895. * Use block scope thread_local static for NSProgress progress stack (#1912) * Use block scope thread_local static for NSProgress progress stack: Works around issue with file_scope thread_local in current clang. Leak check unit test added. * cr feedback: collapse initialization * switch to auto and return type deduction * Reintroduce unique_cf's move ctor/assignment. * Enabling some image drawing tests, moving image drawing to a separate… (#1605) * Enabling some image drawing tests, moving image drawing to a separate file and updating some image file names * Fix bug in sizeWithFont (#1917) NSString+UIKitAdditions sizeWithFont:constrainedToSize: should return the height that the text requires rather than the height of the text that can actually fit in the region that CTFramesetterSuggestFrameSizeWithConstraints returns. * Framework changes as a prestep towards supporting TAEF tests. * Add a common test static lib and move all tests to use TAEF under the covers * Small fixes in UTs for debug. * CoreAnimationTest update - disable unsupported CALayer controls; adds a header and footer table entry; affine transformations update (#1874) * CoreAnimationTest update - disable unsupported CALayer controls; adds a header and footer table entry; affine transformations update CHANGES: 1. Supports the enabling/disabling of layer property controls (greyed out if disabled) 2. Custom table view cells for header and footer added - view and number of layers 3. Affine transformations are now applied collectively, 3D transformation still require it to be centered at origin 4. Cosmetic updates and autolayout corrections - IslandWood autolayout still has some quirks (see CGRect- and CGPoint- cells) 5. All specialized layers extracted from external sources are marked as "not supported" even if render partially Fixes #1598 * Code review feedback * Change test output location * Fix GitHub CI build status reporting * Fix the Unit Test build break on OS X. Fixes #1941. * Augment the drawing tests with TAEF-specific behaviour. * Fix interoperability with TAEF setup/teardown. * Add a TAEF entrypoint and clear the win32 stuff out of the OS X one. * Move UIKit.Xaml DependencyProperty registration out of UIKit.Layer's constructor. (#1949) Fixes #1930. * This change includes the addition of a few new features to UIButton. (#1920) * This change includes the addition of a few new features to UIButton. Some functionality deviates from the reference platform, but we feel that it aligns more closely with the default look and feel of Xaml buttons on Windows because buttons 'look' pressed in more cases with these changes, and there are straightforward ways to opt out of this behavior if it's not desired. **adjustsImageWhenDisabled/adjustsImageWhenHighlighted**: We're adding WinObjC-specific support for adjustsImageWhenDisabled and adjustsImageWhenHighlighted. On the reference platform, this tinting/darkening behavior is ONLY applied to the button's background an foreground images, but on our platform, the tintening/darkening is applied to the entire button's bounds (including its background color), but NOT to its label text. The primary justification for this approach is that there's not a straighforward way to tint Xaml Images on RS1, yet this approach also helps us implement simple SystemButton support and it makes UIButtons appear a bit more 'windows-like' in many cases, since they show more distinct pressed behavior. The behavior is as follows; if adjustsImageWhenDisabled is set to YES, and if no UIControlStateDisabled properties are set on the button, then the entire button contents (aside from its label) receives a lightening effect. Similar behavior is implemented for adjustsImageWhenHighlighted. The way to opt-out of this behavior is to either set a single UIControlStateDisabled/UIControlStateHighlighted property, or to set adjustsImageWhenDisabled/adjustsImageWhenHighlighted to NO. **SystemButton (aka buttonWithType)**: We're adding support for UIButtonTypeSystem, which behaves as follows: - Default background is white. - Receives a default text color of (blue/green) [UIColor colorWithRed:0.0f green:0.47843137f blue:1.0f alpha:1.0f] which applies to all states (unless set otherwise) - Default to UIColor lightTextColor for disabled titleLabelColor **SystemButton - Deviations from the Reference Platform**: - When selected, the entire label background turns the default text color (blue/green) [UIColor colorWithRed:0.0f green:0.47843137f blue:1.0f alpha:1.0f], and the text color is changed to white. This seems like a corner case that we're not implementing until we need to for a customer. - When highlighted, the titleLabel contents (its text and any sublayer content) are tinted to a darker/greyed out version of the normal state. - Instead of this behavior, we're piggy-backing off of our adjustsImageWhenHighlighted implementation to avoid needing to do this extra work. The button *looks* highlighted, so we don't have to change the text color. - On the reference platform, when disabled, text is grey by default, UNLESS any custom normal or disabled text colors are set. On our platform, we DEFAULT to disabled to grey text and then you can change it, but changing the normal color won't affect the default color (for the sake of simplicity). Scrubbed Annotations: All UIButton annotations have been scrubbed and updated with 'NotInPlan' status as needed. Fixes #1442 Fixes #1885 Fixes #1671 Fixes #1671 Fixes VSO 10145069. Fixes VSO 9533114. * Adding comments + clang-format. * Incorporating CR feedback. * Fixing up UIButton tests. * Partial revert of 8bbd30f (#1957) * Partial revert of 8bbd30f * Fixing lab tests. Also combinining tests into 1 common test. (#1951) * Fixing lab tests. Also combinining tests into 1 common test. * CR Feedback. * CR feedback * Projection binary update: Fixed ARC issues caused by names starting with new/copy/alloc * Projections binary update: fixed annotations for non-object types * Working around #1965 in UIImageSetLayerContents. (#1967) Fixes #1962. * Adding support for borders to UILabel and UIButton (via CALayer and the CoreAnimation/UIKit composition layer). Note that we still don't have border support on basic CALayers, because we haven't needed to add them yet, but the plumbing is in now place to add them when needed. (#1956) Fixes #1884. * Supporting gesture subclassing and firing delegates for supporting gestures concurrent firing negotiations. (#1909) * Supporting gesture subclassing and firing delegates for negativating concurrent firing * Comments * Fix some regressions * separate the gesture processing code into its own class and adding initial tests * CR comments * Comments * comments * delete UIGestureRecognizerDelegate.h which causes protocol forward usage which av the tests * Add support for all CGTextDrawingModes (#1933) Several changes have been added to address #1933 and #1935 CGContextDrawGlyphRun has been changed to _CGContextDrawGlyphRuns and __CGContext::DrawGlyphRun -> __CGContext::DrawGlyphsRuns, which takes a dummy struct GlyphRunData, which holds the DWRITE_GLYPH_RUN, relative position of the glyph run (to the text position), and attributes of the glyph run (fill color, etc.). Text drawing methods such as CTFrameDraw now accumulate an array of GlyphRunDatas and pass them to _CGContextDrawGlyphRuns, which in turn calls context->DrawGlyphRuns. To batch all run drawing for clipping/stroking into the same geometry, the runs have to be drawn with the same transformation, so position has to be given (hence relative position in GlyphRunData). The position is transformed by the inverse of the text transform, so when the glyph is drawn and transformed by the real text transform it will have the correct position. This has the added benefit of allowing us to draw a glyph run which has a transformed text matrix all together, rather than glyph-by-glyph as we did previously. Because text drawing is being batched, attributes which are handled at draw time on a run-per-run basis such as fill color need to be done in DrawGlyphRuns. The attribute names are defined in CoreText, though all of them are just CoreGraphics attributes, so local values have been added to CGContext.mm and exported for the CT/UIKit versions as CG. Images have been updates since changing this seems to have enabled antialiasing to some degree on several tests. Fixes #1933 Fixes #1935 * Fixing naming of UIKit.Label helper method: TextBox->TextBlock. (#1970) * Add support for colored and non-colored Patterns (#1952) * Add support for coloured and non-colored Patterns * Enables proper anchoring for patterns * Avoid D2D1Group fill mode conversion for CGPath. (#1954) * Avoid D2D1Group fill mode conversion for CGPath and other simple geometries. Fill mode conversion removed from CGContext. * Remove unnecessary check for proper fill mode since only one fill mode will ever be converted to. * Parameterized test for fill modes with embedded circles with alternating arc directions. * Fix Fill/Stroke tests and remove redundant tests. * Fix relative project path * Updating version for 1702 release
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Creates a new test project Benchmark in the Unit Test group, which leverages GTest to run automated benchmark tests. Simple tests can be done with the TEST_BENCHMARK macro, or more complex ones can be used by creating classes to do setup and teardown so only specific parts of code are measured.
Fixes #1819