Move the initialization of SharedSurfacesParent from the compositor
thread creation to mirror the other WebRender-specific components, such
as the render thread creation. Now it will only be created if WebRender
is in use. Also prevent shared surfaces from being used by the image
frame allocator, even if image.mem.shared is set -- there is no purpose
in allowing this at present. It was causing startup crashes for users
who requested image.mem.shared and/or WebRender via gfx.webrender.all
but did not actually get WebRender at all. Surfaces would get allocated
in the shared memory, try to register themselves with the WR render
thread, and then crash since that thread was never created.
The recovery is only enabled for release builds, to avoid papering over
such hit testing bugs. On nightly builds, a diagnostic assert is issued.
MozReview-Commit-ID: Aos0j0jv6Lb
--HG--
extra : rebase_source : 1cb7fa24ed1f3cba7344f52670626bdeb41b51cc
When ContentChild::RecvInitRendering is received, it tries to setup the
IPDL actors related to rendering. If the GPU process crashes before or
during this process, it will fail, and cause the content process to
crash as well. This is unnecessary because the UI process will either
restart the GPU process, or subsume its job into itself, and trigger
ContentChild::RecvReinitRendering. It is a similar case for failures in
ContentChild::RecvReinitRendering.
Since the GPU process crashing should be a recoverable scenario, we now
check if the remote IPDL actor is in the UI or the GPU process. If it is
in the UI process, it will fail/crash as it does today. If it is in the
GPU process, it will wait for the next
ContentChild::RecvReinitRendering.
For failures that are not IPDL related (e.g. failed to get some resource
like spawning a thread), we release assert specifically for those
failures. They are not recoverable.
This commit introduces a lock on each FilterNodeSoftware around the cached surface.
This lock must be acquired whenever the cached surface is accessed. Currently this
is when the Filter is invalidated or painted.
When painting we only hold the cache lock when determining if we need to compute
a new surface, and after we have computed that new surface. This allows another
thread that may be painting that filter node at a different rect to not contend
on the cache. This is just theoretical though as DrawTargetTiled doesn't clamp
the source rect of the Filter command to the tile rect.
MozReview-Commit-ID: KFVZ35e1czB
--HG--
extra : rebase_source : a22918df80a24fad216a834a5737249cab570f08
This is a just a back out of changeset 6882857e1bb5.
MozReview-Commit-ID: 9Z6iEHl3eAy
--HG--
extra : rebase_source : 3d091c72a6e589e702d90ffe6800e131b009604b
The image decoding thread pool can grow to be quite large, up to 32
threads, depending on the number of processors on the system. If the
user is not actively browsing, these threads are occupying resources
which could be reused elsewhere. After the timeout period, it will
release up to half of the threads in the pool.
We will re-use these methods to perform various calculations once we introduce
the pending playback rate.
MozReview-Commit-ID: 2HV44TTNxHg
--HG--
extra : rebase_source : 19e32b13c6fee9cfe9993918557f2433aa687e11
The change to RootAccessible.cpp fixes an obvious bug introduced in bug 741707.
The visibility changes in gfx/thebes are because NS_DECL_ISUPPORTS has a
trailing "public:" that those classes were relying on to have public
constructors.
MozReview-Commit-ID: IeB8KIJCGhU
There's only one caller so let's just eliminate inline FillAzure into
Fill.
MozReview-Commit-ID: JYpAQMkhEcS
--HG--
extra : rebase_source : 962b046405949d435537a77baea50f631a8aee72
Add a command CreateClippedDrawTarget to DrawTarget, which takes the
max required size and a transform between this draw target and the one
to be created. The created draw target may have its size clipped to
the size of this draw target, transformed to the new target's
space. This means that the new surface will be large enough so
that it is rendered to this draw target correctly, but not necessarily
any larger.
Usually this will just create a draw target of the requested size, for
simplicity. However, when replaying a recorded draw target we do clip
the size to the base draw target's size. This is done using a
DrawTargetTiled, so when applying the mask in PopLayer, we must take
the SourceSurface's offset in to account.
MozReview-Commit-ID: 89ONElphzLu
--HG--
extra : rebase_source : 7eebeb66a2686a7b6f4ade36f3004ebb06abc2fe
Fix the DrawTargetTiled::mRect initialization, and expose through the
virtual function GetRect(). Make SnapshotTiled::GetDataSurface()
allocate a surface the size of this rect, rather the XMost and
YMost. Callers of SourceSurface::GetDataSurface() can query
SourceSurface::GetRect() and apply the necessary offset themselves.
MozReview-Commit-ID: C31FGirQ0oK
--HG--
extra : rebase_source : ac31ae3ca0a0b188f9293c4e6898b4e4a65cad0e
Changing the transaction id allocator is now handled better in
ClientLayerManager than before in that it resets the transaction id on
the new allocator to match the old allocator and avoid discontinuities
in the transaction ids. We should apply this behaviour to
WebRenderLayerManager as well, because WebRenderLayerManager was
assuming that any time the allocator changed it would automatically
start the transaction id at 1, which is not the case. In particular,
when navigating to something in the bfcache, we can reuse a pre-existing
refresh driver which might have a transaction id already greater than 1.
MozReview-Commit-ID: 8IUn1Dhnh7c
--HG--
extra : rebase_source : 0e98c96d9636c3a3ee0489ff6b6161bce7677dd7
bug 1379920 changed the way of dirty flag handling. And the dirty flag was not set when CopyableCanvasRenderer::ClearCachedResources() is called. Then ShareableCanvasRenderer::UpdateCompositableClient() did not update canvas.
The important part here is that the test parent page
(test_bug1151663.html) waits for paints to complete before spawning the
helper window. Otherwise, the helper window might think it's done
painting (because it is) and start running the test even though its
layer tree has not been attached to the chrome layer tree. In such a
scenario reading the APZ test data produces an empty tree for the
content layers id and causes the test to fail.
This test was written before we had waitUntilApzStable so sprinkling some
of that into the test makes it more robust in general as well.
MozReview-Commit-ID: J8rqW6dcy23
--HG--
extra : rebase_source : f9e28bd15f355c88649ff1797b1f1205824bf815
A Box is like a Rect but represented as (x1, y1, x2, y2) instead of
(x, y, w, h).
The API is bare-bones at the moment; it can be extended as needed
by future users.
MozReview-Commit-ID: FWv69Y5hP6t
--HG--
extra : rebase_source : 1f717727bc724842a2f6adcba9e6cbbe50059436