This patch addresses a number of inconsistencies in the device initialization process, as well as simplifying it for future use. All device decisions are now explicitly made up-front during startup, rather than implicitly or on-demand. In addition a number of restrictions have been placed on when we can construct devices.
Full change list:
(1) We no longer attempt to use D3D11 if acceleration is disabled or D3D9 is preferred. This is a departure from our previous behavior, where we would construct these devices but then not use them as a compositor backend.
(2) D3D11 startup no longer creates a content device (this is reserved for D2D initialization).
(3) D2D is only attempted if we managed to create a D3D11 compositor device. This is a departure from previous behavior where if D3D11 was not used for compositing, we could still create its machinery to use D2D as a content backend.
(4) D2D 1.1 initialization is now directly responsible for creating a D3D11 content device.
(5) D2D 1.0 and 1.1 logic have been disentangled for clarity.
(6) UpdateRenderMode() has been split up, so we can update backend prefs out of band with device resets.
(7) mUseGDIFonts and mUseDirectWrite have been removed as their state was confusing. Instead, D2D now depends on DWrite initialization succeeding. If later we fail to get a DWrite font list, we revert our decision to use Direct2D.
(8) Device resets now clear a little more state, including the devices set in Moz2D Factory.
(9) We no longer create a DWrite text analyzer as it was unused.
This conversion was done with the script:
find . -name '*.cpp' -o -name '*.h' -o -name '*.mm' -o -name '*.idl' | \
egrep -v 'cairo-win32-refptr.h|RefPtr.h|TestRefPtr.cpp' | \
xargs sed -i -e 's/mozilla::TemporaryRef</already_AddRefed</g' \
-e 's/TemporaryRef</already_AddRefed</g'
Manual fixups were performed in the following instances:
- We handled mfbt/RefPtr.h manually so as to not convert TemporaryRef itself
into already_AddRefed.
- The following files had explicit Move() calls added to make up for the lack
of a copy constructor on already_AddRefed:
dom/base/ImageEncoder.cpp
dom/media/MediaTaskQueue.{h,cpp}
dom/media/webaudio/PannerNode.cpp
- A redundant overload for MediaTaskQueue::Dispatch was deleted.
- A few manual fixups were required in mfbt/tests/TestRefPtr.cpp.
- Comments, using declarations, and forward declarations relating to
TemporaryRef in dom/canvas/ and gfx/layers/ were changed to refer to
already_AddRefed.
This will hopefully help us track down large amounts of write-combined
mappings that we see.
try: -b do -p win32,win64 -u all -t none
--HG--
extra : rebase_source : 612223046a34a0e7ba179fca4fef1d4c79e18ee8
Bug 1107299 broke this blocking because it assumed that DoesD3D11DeviceWork()
only described whether texture sharing works. However we were also checking for
broken DisplayLink drivers in that function. This splits out that part and uses
it appropriately.
--HG--
extra : rebase_source : 895d2da6297a48d0a2cbef2d149c27471c788484
Calling AcquireSync on textures created with
D3D11_RESOURCE_MISC_SHARED_KEYEDMUTEX appears to be really slow, this really
bites us with the number of textures involved when tiling.
I've also tested using ID3D11Query to check when work is completed, this also
appears to be pretty slow.
This instead just uses a single texture with a keyedmutex, and makes sure we
draw to it last and lock it first. It's pretty hacky, but seems to work really
well so far.
--HG--
extra : rebase_source : bf5f56751f993e507fcd1e5c386fb1fc9a3f73af
Matt Woodrow
David Anderson
Carsten "Tomcat" Book
Bas Schouten
Ryan VanderMeulen
Lee Salzman
Nathan Froyd
Phil Ringnalda
Andrea Marchesini
Chris Pearce ext:(%2C%20Matt%20Woodrow%20%3Cmwoodrow%40mozilla.com%3E)
Ehsan Akhgari
Mason Chang
Jeff Muizelaar
John Daggett
Jonathan Watt
Benoit Jacob