Otherwise with backdrop filters we create and throw away a single fullscreen texture every frame. Very wasteful!
Does not impact iOS because it has the read from resolve option. Removed some unused parts of the inline_pass_context that were only used in the entity pass days.
Fixes https://github.com/flutter/flutter/issues/156361
cc @lyceel. A bit of miscommunication on the priority of this one. I
thought the default font factories were sufficient for the next release.
But it turns out that was not true on the board which is a custom
platform.
In any case, this should unblock custom font registrations. See the
`CanCreateParagraphsWithCustomFont` case for an example. If you don't
provide the family name alias for the custom font, the name will be read
from the font data. This example uses the custom "WhatTheFlutter" custom
font family in "wtf.otf".
<img width="1136" alt="Screenshot 2024-10-17 at 1 26 36 PM"
src="https://github.com/user-attachments/assets/dbbef2fd-f854-4330-8c82-fc8378489769">
Reverts: flutter/engine#55939
Initiated by: jonahwilliams
Reason for reverting: failing on scenario test
Original PR Author: jonahwilliams
Reviewed By: {matanlurey}
This change reverts the following previous change:
Creating descriptor pools is expensive, no need to have more than one per frame.
The lifecycle of the descriptor pool is exactly the same as the cmd pool, which indicates that we can probably do some consolidation/refactoring in the future.
Fixes https://github.com/flutter/flutter/issues/157115
Creating descriptor pools is expensive, no need to have more than one per frame.
The lifecycle of the descriptor pool is exactly the same as the cmd pool, which indicates that we can probably do some consolidation/refactoring in the future.
Fixes https://github.com/flutter/flutter/issues/157115
Introduces a mechanism to allow backdrop filters to 1) share backdrop inputs and 2) fuse filter applications for faster blurs.
This is a proposed solution to https://github.com/flutter/flutter/issues/131568
Implemented:
* Developer can specify a "backdrop id" which indicates that a backdrop layer should share the input texture and potentially cached filter for a layer.
* Removes second save layer for each backdrop filter
* Removes save layer trace event for backdrop filter
* Can fuse backdrop filters if there is more than one identical filter
TBD:
* Adjust heruristic to avoid applying bdf filter to entire screen
Suggestions: applying a bdf should be a distinct operation from a save layer in the DL builder/dispatcher. The saveLayer implmenentation in the impeller dispatcher is super convoluted because it needs to handle both.
### Video
Video starts with normal bdf then I hot reload to specify that the bdfs share inputs/filters. This is running on a pixel 8 pro
Change to the macrobenchmark app is just:
```dart
Widget build(BuildContext context) {
Widget addBlur(Widget child, bool shouldBlur) {
if (shouldBlur) {
return ClipRect(
child: BackdropFilter(
filter: ImageFilter.blur(sigmaX: 5, sigmaY: 5),
backdropId: 1, // Added ID
child: child,
),
);
} else {
return child;
}
}
```
https://github.com/user-attachments/assets/22707f97-5825-43f1-91b4-1a02a43437f5
Requires framework changes in https://github.com/jonahwilliams/flutter/pull/new/backdrop_id
Avoid Sprintf'ing labels for textures/cmd buffers/render passes. For the render target textures, defer combined label construction until we know the label will be used.
From local testing:
1. This is slightly faster
2. This still renders correctly even on old phones (S10)
3. This is validation error clean.
---------------
The vk RenderPass has a built in mechanism to transition the attachments between layout states. Since we expect the images to be in eGeneral layout, we can configure the render pass to do an eUnderfined -> eGeneral transition. eUnderfined means "we don't care what was here before". This is OK because we're about to clear the texture anyway.
Since the RenderPass is doing the transition, we don't need an explicit barrier. All we have to do is update our own state tracking mechanism that the image attachments should behave as if they were in eGeneral layout - because they are once the render pass is started.
Many of our labeling APIs accepted a std::string. When provided with a string literal (const char*), this would force the allocation of a new std::string object and a copy of the original char*. std::string_view is a view of either a std::string, or a const char* (or other), which allows an API that doesn not need to store the string to accept both without a copy. Of course, we can't store a std::string_view, as that doesn't provide memory ownership - so many of the GLES implementations will still copy the string_view into a string locally.
Also updates several of the debug labelling functions to no-op in release mode and removes some expensive interpolations that aren't super necessary.
Remove the extra host -> host copy performed by the VertexBufferBuilder. Instead, write the drawPoint geometry right into the transients buffer.
https://github.com/flutter/flutter/issues/152702
This is the beginning of the bulk of de-skia-fication work in the engine. All of the standard types in the DlCanvas API now have overloads that specify the corresponding Dl type, mainly for Dl*Rect and DlPoint types. This enables further work to switch from SkFoo types to DlFoo types in the various engine modules culminating in the elimination of the old methods that use the Sk types.
All of the former methods that used the basic Sk types are now implemented as inlinable translation overloads and the underlying implementations of DlCanvas now implement only the newer style interfaces so that they don't need to be further modified as we eliminate the old Skia types from the interface.
There are still a couple of Skia types remaining in the DlCanvas API without any DL type variants which will be handled in a future phase:
- SkRRect
- SkRSXform
- SkTextBlob (will be hidden behind a common interface along with TextFrame)
- SkImageInfo (only used in a few calling sites)
Almost all geometry objects can now be stack allocated while rendering, instead of heap allocated - due to the removal of the EntityPass compositor. Clip geometries must still be heap allocated as the geometry objects are not yet stored in the display list, and the clips must be recorded for backdrop filter clip replay.
The canvas stores unique_ptrs to these geometries in a vector that it clears at the end of the frame. The Geometry factory methods were adjusted to return unique ptrs, and the Geometry class given a virtual dtor since we no longer rely on the special property of std::shared_ptr for memorizing the right dtor.
The ColorSourceContents and ClipContents class now hold onto a const ptr to the geometry object.
At some point in the future, we can rework the geometry object to be stored inline in the display list which will further simplify this code.
Part of https://github.com/flutter/flutter/issues/142054
b/371512414
This issue had no reproduction but the stacktrace was provided. The theory of this fix is that the `ImageResult` type is capturing something that doesn't handle being copied and deallocated twice so instead of copying it, we std::move it into a shared_ptr so that it will only be deallocated once.
Also, I just avoid using a `std::move` when overflowing. The idea here is maybe sometimes deleting the std::move'd item isn't kosher. We know from the stacktrace that it has something to do with overflowing because that would be the only case where something is being deleted.
I've added 2 integration tests that exercises the code where the crash appears to be happening. This is good coverage to have but neither of them reproduced the issue, even with MallocScribble enabled. That's the best we can do without a reproduction.
```
Thread 7 (id: 0x0000a103)crashed
0x000000010a48ae8c(Flutter -function.h)std::_fl::allocator<impeller::ContextMTL::PendingTasks>::destroy[abi:v15000](impeller::ContextMTL::PendingTasks*)
0x000000010a48b9fc(Flutter -allocator_traits.h:309)impeller::ContextMTL::StoreTaskForGPU(std::_fl::function<void ()> const&, std::_fl::function<void ()> const&)
0x000000010a48b9fc(Flutter -allocator_traits.h:309)impeller::ContextMTL::StoreTaskForGPU(std::_fl::function<void ()> const&, std::_fl::function<void ()> const&)
0x000000010a4d49e4(Flutter -image_decoder_impeller.cc:422)std::_fl::__function::__func<flutter::ImageDecoderImpeller::UploadTextureToPrivate(std::_fl::function<void (sk_sp<flutter::DlImage>, std::_fl::basic_string<char, std::_fl::char_traits<char>, std::_fl::allocator<char>>)>, std::_fl::shared_ptr<impeller::Context> const&, std::_fl::shared_ptr<impeller::DeviceBuffer> const&, SkImageInfo const&, std::_fl::shared_ptr<SkBitmap> const&, std::_fl::optional<SkImageInfo> const&, std::_fl::shared_ptr<fml::SyncSwitch> const&)::$_1, std::_fl::allocator<flutter::ImageDecoderImpeller::UploadTextureToPrivate(std::_fl::function<void (sk_sp<flutter::DlImage>, std::_fl::basic_string<char, std::_fl::char_traits<char>, std::_fl::allocator<char>>)>, std::_fl::shared_ptr<impeller::Context> const&, std::_fl::shared_ptr<impeller::DeviceBuffer> const&, SkImageInfo const&, std::_fl::shared_ptr<SkBitmap> const&, std::_fl::optional<SkImageInfo> const&, std::_fl::shared_ptr<fml::SyncSwitch> const&)::$_1>, void ()>::operator()()
0x000000010a06c234(Flutter -function.h:512)fml::SyncSwitch::Execute(fml::SyncSwitch::Handlers const&) const
0x000000010a4d42f8(Flutter -image_decoder_impeller.cc:408)flutter::ImageDecoderImpeller::UploadTextureToPrivate(std::_fl::function<void (sk_sp<flutter::DlImage>, std::_fl::basic_string<char, std::_fl::char_traits<char>, std::_fl::allocator<char>>)>, std::_fl::shared_ptr<impeller::Context> const&, std::_fl::shared_ptr<impeller::DeviceBuffer> const&, SkImageInfo const&, std::_fl::shared_ptr<SkBitmap> const&, std::_fl::optional<SkImageInfo> const&, std::_fl::shared_ptr<fml::SyncSwitch> const&)
0x000000010a4d3838(Flutter -image_decoder_impeller.cc:538)std::_fl::__function::__func<flutter::ImageDecoderImpeller::Decode(fml::RefPtr<flutter::ImageDescriptor>, unsigned int, unsigned int, std::_fl::function<void (sk_sp<flutter::DlImage>, std::_fl::basic_string<char, std::_fl::char_traits<char>, std::_fl::allocator<char>>)> const&)::$_1, std::_fl::allocator<flutter::ImageDecoderImpeller::Decode(fml::RefPtr<flutter::ImageDescriptor>, unsigned int, unsigned int, std::_fl::function<void (sk_sp<flutter::DlImage>, std::_fl::basic_string<char, std::_fl::char_traits<char>, std::_fl::allocator<char>>)> const&)::$_1>, void ()>::operator()()
0x000000010a06dd1c(Flutter -function.h:512)fml::ConcurrentMessageLoopDarwin::ExecuteTask(std::_fl::function<void ()> const&)
0x000000010a06737c(Flutter -concurrent_message_loop.cc:101)void* std::_fl::__thread_proxy[abi:v15000]<std::_fl::tuple<std::_fl::unique_ptr<std::_fl::__thread_struct, std::_fl::default_delete<std::_fl::__thread_struct>>, fml::ConcurrentMessageLoop::ConcurrentMessageLoop(unsigned long)::$_0>>(void*)
0x000000021d4ff378(libsystem_pthread.dylib + 0x00006378)_pthread_start
```
[C++, Objective-C, Java style guides]: https://github.com/flutter/engine/blob/main/CONTRIBUTING.md#style
Fixes https://github.com/flutter/flutter/issues/155877
Technically SurfaceControl is supported on API 29 but I've observed enough reported bugs that I'm bumping the constraint to 30. If we had more time to test all of these older devices maybe we could figure out if the problem is something that can be worked around.
The Vulkan command buffer wrapper delegates almost all of its functionality to the "CommandEncoderVK". I did not find that this separation was useful, as they have a 1-1 relationship and an identical lifecycle. Lets combine them to reduce the number of things to worry about.
We will need to denylist certain Vulkan drivers. This is easier to do if we have some mechanism of determining which GPU model we're currently using. This doesn't add any new denylisting, it just ads a mechanism to parse the driver name into an enum. I mean, there are only so many SoCs, right?
See also https://github.com/flutter/flutter/issues/155185
Use a Squircle Signed Distance Field based algorithm to do a very fast approximation of rrect blurs.
This PR is to provide reference within the team to discuss the future of the algorithm compared to the Gaussian approximation functions that are currently in use. It isn't a complete solution, but can be completed easily with a little more work.
Notably, it doesn't handle elliptical round rects, only circular corners.
Could stand to include an attribution to the source (https://raphlinus.github.io/graphics/2020/04/21/blurred-rounded-rects.html)
Aiks color source is more or less a one to one copy of the impeller::DisplayListColorSource. We can rip it out, instead of flutter::DisplayList::ColorSource -> impeller::Aiks::ColorSource -> impelller:Entity::ColorSourceContents, just go directly flutter::DisplayList::ColorSource -> impelller:Entity::ColorSourceContents.
Part of https://github.com/flutter/flutter/issues/142054
Aiks has been gutted, and most of the reming types need to be merged with dl/entity types and deleted. this is difficult to do with impeller/display_list and aiks as separate gn targets as it requires circular deps.
Lets merge them and remove the aiks directory.
