When the Android runtime starts, the system preloads a series of assets
in the Zygote process. These assets are shared across all processes.
Unfortunately, each one of these assets is later uploaded in its own
OpenGL texture, once per process. This wastes memory and generates
unnecessary OpenGL state changes.
This CL introduces an asset server that provides an atlas to all processes.
Note: bitmaps used by skia shaders are *not* sampled from the atlas.
It's an uncommon use case and would require extra texture transforms
in the GL shaders.
WHAT IS THE ASSETS ATLAS
The "assets atlas" is a single, shareable graphic buffer that contains
all the system's preloaded bitmap drawables (this includes 9-patches.)
The atlas is made of two distinct objects: the graphic buffer that
contains the actual pixels and the map which indicates where each
preloaded bitmap can be found in the atlas (essentially a pair of
x and y coordinates.)
HOW IS THE ASSETS ATLAS GENERATED
Because we need to support a wide variety of devices and because it
is easy to change the list of preloaded drawables, the atlas is
generated at runtime, during the startup phase of the system process.
There are several steps that lead to the atlas generation:
1. If the device is booting for the first time, or if the device was
updated, we need to find the best atlas configuration. To do so,
the atlas service tries a number of width, height and algorithm
variations that allows us to pack as many assets as possible while
using as little memory as possible. Once a best configuration is found,
it gets written to disk in /data/system/framework_atlas
2. Given a best configuration (algorithm variant, dimensions and
number of bitmaps that can be packed in the atlas), the atlas service
packs all the preloaded bitmaps into a single graphic buffer object.
3. The packing is done using Skia in a temporary native bitmap. The
Skia bitmap is then copied into the graphic buffer using OpenGL ES
to benefit from texture swizzling.
HOW PROCESSES USE THE ATLAS
Whenever a process' hardware renderer initializes its EGL context,
it queries the atlas service for the graphic buffer and the map.
It is important to remember that both the context and the map will
be valid for the lifetime of the hardware renderer (if the system
process goes down, all apps get killed as well.)
Every time the hardware renderer needs to render a bitmap, it first
checks whether the bitmap can be found in the assets atlas. When
the bitmap is part of the atlas, texture coordinates are remapped
appropriately before rendering.
Change-Id: I8eaecf53e7f6a33d90da3d0047c5ceec89ea3af0
Merge simple bitmap draw operations and text operations to avoid
issuing individual gl draws for each operation. Merging other ops to
be done eventually.
The methods are different - the bitmap merging generates a single
mesh for reused, unclipped images (esp. repeated images in a listview)
The text approach queries just defers the normal font rendering until
the last drawText in the sequence that can share the same shader.
Patches are sorted and merged, but don't yet have a multiDraw
implementation. For now, the pretending-to-merge gives better sorting
behavior by keeping similar patches together.
Change-Id: Ic300cdab0a53814cf7b09c58bf54b1bf0f58ccd6
bug:8501661
Makes the performance issues and interaction with layer type/paint
more clear.
Additionally, corrects change from 47ab7d6612e2b5b8b66fb261dafef7c91264e173
to still allow displayList alpha to override layer paint
Change-Id: Ic94d75865700820489370461cd8ac9f9077a8d90
bug:8464795
Changes drawModifiers and alpha to be restored for all operations,
since saveLayer/restore use these values, not just draw operations
Also forces a renderer state restoration before a deferred restore op
is played back, in case it is associated with a saveLayer that
doesn't have the save_clip flag set
Change-Id: I9da5d44fefbfffdee164c98f4f139843dacf85df
bug:8471552
Additionally, add ignore check for draw functor
In deferred mode, if we don't set ignore, a 0 alpha displayList won't
be quickRejected, and not only would its contents be needlessly
deferred, but upon playback any functors would be drawn, ignoring the
state of the snapshot.
Change-Id: Ib1edb3252080f674254086edee6f965f0efcce78
Bug #8378964
This change defers drawing into layers until after the renderer for FBO0
is ready to draw. At that point, all the precaching is done which means
all glyphs can be uploaded at once in the font caches.
Change-Id: Ie1f7a7ff30f76f06fb3dbc72c7d05e66207d1ecb
bug:8409891
Snapshots frequently have their clip overwritten due to applying
deferred state - now, store tiling clip information in a separate
rect, outside of the snapshot so it isn't overwritten.
Change-Id: I21ca4c45dcd802eae99e8de86f11525196777ccb
Only performs the GL setup steps when at least one glyph is drawn.
This change also skips various draw operations when the specified
paint draws with alpha = 0.
Change-Id: I9eda148b0503acffc552ee19196f5d52e958a1a2
bug:8409891
In order to defer fbo-targetting saveLayer operations, it's necessary
to update the snapshot at defer time so that deferred display state
(namely, clip and transform) are fbo relative.
Re-enables deferring, as the issues with saveLayer are fixed.
Change-Id: I74b0779bc732675c747208f0757c3ea85f6dfbed
bug:8401910
This avoids leaving the renderer in a state with stale pointers (to,
e.g., shaders or color filters)
Change-Id: Idf8b63657041352e70e34e91ea416fe4385d4bc2
Removes mMultipliedAlpha, using the snapshot alpha for all
non-overlapping display list alpha control.
Additionally, fixes opacity issues where children of
hasOverlappingRendering=false displaylists (both hw layer sublists and
other sublists with hasOverlappingRendering=false)
Change-Id: I6adc16da855835f9f518f8967628e5d0135c789b
bug:8037003
A recursive drawDisplayList call is now entirely deferred before
playing back to the screen and issuing GL commands. This way, the
entire stream can be inspected, optimized, and batch work (such as
uploading textures) before issuing commands.
Additionally, this fixes an issue where operations draw could move
across restores corresponding to saveLayer(alpha). Those and other
similar cases (such as complex clipping, requiring the stencil) are
now treated as batching barriers, with the operations that change
renderer state in a way that's difficult to defer are just re-issued
at flush time.
Change-Id: Ie7348166662a5ad89fb9b1e87558334fb826b01e
This change extracts the scale parameters of the current transform
to pass then to the font renderer. Rotation and perspective are
applied to the generated mesh inside the vertex shader. This limits
the number of glyphs we have to create in the font cache and thus
reduces memory churn.
Change-Id: Ic5b3bae2b2b0e0250a8ee723b071a1709725c749
The drawColorRects() method was clipping individual rectangles
using the wrong parameters left, top, right and bottom instead
of l, r, t and b. It also checked for count == 0 after the loop
when it should have checked for vertexCount == 0. The quickReject
is now not part of the loop since it's a bit overkill to perform
so many matrix multiplications. What we really care about is the
final quickReject performed on the max bounds of the entire set
of rectangles.
This change also replaces all instances of mSnapshot->transform
by currentTransform() to make the code slightly more readable.
Change-Id: I6485280414499716852f7dbfba186774eb6763d4
The deferred display lists model now allows us to precache glyphs
at their exact size on screen.
This change also removes debug markers when the renderer defers
and reorders display lists. It also adds a flush event marker.
Change-Id: I66ec5216dc12b93ecfdad52a7146b1cfb31fbeb4
This change does not apply to drawPosText() and drawTextOnPath() yet.
Prior to this change, glyphs were always rasterized based on the
font size specified in the paint. All transforms were then applied
on the resulting texture. This creates rather ugly results when
text is scaled and/or rotated.
With this change, the font renderer will apply the current transform
matrix to the glyph before they are rasterized. This generates much
better looking results.
Change-Id: I0141b6ff18db35e1213e7a3ab9db1ecaf03d7a9c
When the renderer draws a bitmap as part of a display list with an
alpha < 1.0f, the paint is temporarily modified to alter the opacity
of the bitmap. This has the side effect of increasing the paint's
generation ID counter which can break paint caching.
Change-Id: I5055d059ad1639829fa50af3d946e296c4dab877
bug:8235699
Ensure rectangle clipping operations disable deferring when necessary
(i.e., when the op might create a non-rect region), including in
DisplayList::setViewProperties
Additionally, makes clipping with a kUnion always use a region, for
consistency with software rendering
Change-Id: I6730f1a80250bcf3f91cd4afde646d470a12dbc2
bug:8037003
The reordering enables similar operations to draw together, minimizing the
OpenGL state change operations that go inbetween draws. Eventually, multiple
complete canvas draw operations will be merged (into a single glDrawArrays call,
for example)
Reorders DisplayList draw operations when:
-They can move backwards in the command stream to be after similar
operations without violating draw ordering
-The OpenGLRenderer is in a simple, replayable state (no complex clip,
or filter/shadow etc)
Also adds two system properties to control the deferral/reordering:
"debug.hwui.disable_draw_defer"
"debug.hwui.disable_draw_reorder"
which can be set to "true" to control the display list manipulation
Change-Id: I5e89f3cb0ea2d2afd3e15c64d7f32b8406777a32
Bug #7146141
This change is needed to add a render buffer cache to avoid
creating and destroying stencil buffers on every frame.
This change also allows the renderer to use a 1 bit or 4 bit
stencil buffer whenever possible.
Finally this change fixes a bug introduced by a previous CL
which causes the stencil buffer to not be updated in certain
conditions. The fix relies on a new optional parameter in
drawColorRects() that can be used to avoid performing a
quickReject on rectangles generated by the clip region.
Change-Id: I2f55a8e807009887b276a83cde9f53fd5c01199f
bug:7117155
bug:8114304
Currently used for lines (with and without AA) and arcs with useCenter=false
Also removes 0.375, 0.375 offset for AA lines
Change-Id: Ic8ace418739344db1e2814edf65253fe7448b0b0
Bug #7146141
The GL_QCOM_tiled_rendering extension requires careful use of
start/endTiling when attaching a renderbuffer dynamically.
Change-Id: I20036683ed3909ffaf40cc3d57a25257e35b6fa2
Bug #7146141
When non-rectangular clipping occurs in a layer the render buffer
used as the stencil buffer is not cached. If this happens on a
View's hardware layer the render buffer will live for as long
as the layer is bound to the view. When a stencil buffer is
required because of a call to Canvas.saveLayer() it will be allocated
on every frame. A future change will address this problem.
If "show GPU overdraw" is enabled, non-rectangular clips are not
supported anymore and we fall back to rectangular clips instead.
This is a limitation imposed by OpenGL ES that cannot be worked
around at this time.
This change also improves the Matrix4 implementation to easily
detect when a rect remains a rect after transform.
Change-Id: I0e69fb901792d38bc0c4ca1bf9fdb02d7db415b9
Region clipping, using Canvas.clipPath or Canvas.clipRegion, requires
a stencil buffer to be always present. In addition, extra wiring is
required in JNI and display lists.
This change only adds the necessary JNI/C++ APIs and some extra
plumbing to start the real work on properly supporting region
clipping.
A new debug define called DEBUG_CLIP_REGIONS can be used to draw
the current clip region. It is off by default, as is region
clipping.
The default implementation of clipPath() and clipRegion(), now
in native, mimics the previous Dalvik implementation to prevent
regressions.
Change-Id: I7903e7cfd7412b9b9b622566d4dbfce7bdcec00c
When profiling is enabled with debug.hwui.profile set to true,
setting debug.hwui.profile_visualizer to true will display the
profiling data directly on screen.
Change-Id: I3d5fe3f0347090815087b1cbfce66b8e76d9347b
Bug #7326824
When a layer is taken out of the cache and initialized it gets cleared
to prepare it for future rendering. This triggers the following sequence
of operations:
glBindFramebuffer(layer.fbo)
attach texture storage to FBO
glClear()
glBindFramebuffer(defaultFbo)
The clear forces a resolve on tilers which stalls the CPU for a little
while, thus producing jank during animations. This change moves the
clear to the next frame when we know we will have to execute a resolve
anyway.
Change-Id: Ic1939c25df20ed65a4c48dc81ee549b2cd8b6ec3
* changes:
Skia's ColorMatrix vector is in the 0..255 range not 0..1 Bug #7248980
Don't use the QCOM_tiled_rendering extension with functors Bug #7247880
bug:4419017
bug:7230005
- Adds support for stroke/strokeAndFill for shapes without joins
- Fixes path-polygonization threshold calculation
- Fixes rendering offset (now only used for points)
- Several formatting fixes
Change-Id: If72473dc881e45752e2ec212d0dcd1e3f97979ea
Bug #7158326
When scissor optimization is enabled, OpenGLRenderer will attempt to
minimize the use of scissor by selectively enabling and disabling the
GL scissor test.
When the optimization is disabled, OpenGLRenderer will keep the GL
scissor test enabled and change the scissor rect as needed.
Some GPUs (for instance the SGX 540) perform better when changing
the scissor rect often than when enabling/disabling the scissor test
often.
Change-Id: Idb68862e287a23358f9188d577ae0f86161902fd
Bug #7186819
This optional OpenGL extension can be used by tiled renderers to optimize
copies from main memory to tiles memory.
Change-Id: Id4a5d64e61ad17f50e773e8104b9bf584bb65077
