Age | Commit message (Collapse) | Author |
|
This reverts commit 8fdfdeb01654604d429785c4216373449e516b9d.
Reason for revert: breaks emulator
BUG=255449023
Change-Id: I61917f1c8ca3d003bc70e8f31947aa00039864d0
Merged-In: I0ec7e0ba6ec9aeb5a23f20391b8524f87e4f343b
|
|
This updates vulkan-cereal to aosp/master.
BUG=251519884
TEST=run Cuttlefish with gfxstream
Change-Id: I1128ad6cbfc0dc865248d903ecb9fa4ac3931abb
|
|
|
|
|
|
|
|
Bug: b/246990929
Change-Id: Icf1726cc3b17b331949ee14caf4581ef66e1cb61
|
|
Bug: 237805597
Test: presubmit
Change-Id: I0ec7e0ba6ec9aeb5a23f20391b8524f87e4f343b
|
|
|
|
|
|
This reverts commit 96bc3be281ebbcbbdcbaa6108054f0a2925022e1.
After the original change, during swapchain recreation, the Windows
downstream either hits device lost or hangs.
Change-Id: I1cf2ffa6e3185a823e70fb8dafe6e9899c6b9932
|
|
Test: build
Change-Id: I642f2896b86f29e5f773d09777e85adefef766c4
|
|
Change-Id: Id575c78d8c5760d4edc39fd7847745636ba2d472
|
|
On my machine, this results in a x5 speedup with 8 threads. Average decompression speed goes from 25 to 126 Mpixels/s for Summoners Wars: Chronicle.
This also makes ASTC CPU decompression 3x faster than our current compute shade approach. I measured the loading times of the Aztec Ruins benchmark, which uses a lot of ASTC decompression before starting:
- Compute shader decompression: 32 seconds
- Multi-threaded CPU decompression: 10.5 seconds
Change-Id: Id0fea5e3287231f13ad05ddbfc9bcc2a3bee5c1f
|
|
|
|
...to monitor specifically and to prevent spurious hangs
as a result of long decompression.
Test: Built and ran emulator
Bug: 247865035
Change-Id: Ia553acd733f8c3449e5151d31bbb8a70cff3db89
|
|
This reverts commit cb8d1e4f237be8d58e26de57ee0b67245a63a34f.
Reason for revert: Seeing issues with potential concurrent usage of the pbuffer context.
Bug: b/251366833
Bug: b/250847245
Change-Id: I39ba968339009ba8086b5f0c989aa5ba50f5a9aa
|
|
|
|
Bug: 237805597
Test: presubmit
Change-Id: I53c00f6cea9dcf63f9725673e7f721e79e8b727c
|
|
|
|
|
|
Change-Id: I794db8ad0d0230ca59ad69e8e1aad74f73eddb0f
|
|
using an intermediate buffer.
Change-Id: Icdf7e614f225a3d9c248583aa34e771a12653e84
|
|
|
|
Bug: 237805597
Test: presubmit
Change-Id: If74accd3f7ae23e095132fada79973024ecf0211
|
|
... to avoid using the ColorBuffer context helper across multiple
threads (virtio gpu thread and post worker thread). This issue
was accidentally introduced in aosp/2232570.
Bug: b/250847245
Test: build
Change-Id: If46100eb369a040c5e9bf2db1e0bc203ad3f6656
|
|
Change-Id: Icb49eb5ccfb31d4bf5a352e36dcad77dcde101d8
|
|
This breaks down the AstcCpuDecompressor class into 2 parts:
- *AstcCpuDecompressor*
- Now moved to stream-servers/compressedTextureFormats/
- Purely concerned about ASTC decompression. Doesn't call any Vulkan APIs.
- *AstcTexture*
- Helper class used in VkDecoderGlobalState to handle a ASTC-compressed VkImages
- Holds state (ie: was decompression successful?) and resources (VkBuffer, etc..)
There should be no logic changes compared to the old code.
This CL sets the stage for multi-threaded ASTC decompression, and allows the ASTC decompression code to be shared with others.
Change-Id: I701c83d6b24aa71017a131d44c78e6118c4dcd93
|
|
|
|
Adds driver and device UUID, deprecates physical_device_idx.
Bug: b/244622199
Test: downstream
Change-Id: I1f8cce09607731d95404b7a0c0df2aaa47a4806d
|
|
|
|
Bug: b/250847245
Test: build
Change-Id: I1df857a2423d79ed81ee5d789ae0e782e98dfa4d
|
|
This change moves AstcCpuDecompressor to `compressedTextureFormats/`.
This sets the stage for the second step where we will break this class into 2 classes, and hopefully makes reviewing the change a little bit easier.
Change-Id: I0e4eb29b0de94ac599ec4f4034c22dcb41c2cd0c
|
|
Caught by the static analyzer:
device/generic/vulkan-cereal/stream-servers/vulkan/VkCommonOperations.cpp:1235:48:
warning: Access to field 'ivk' results in a dereference of a null
pointer (loaded from variable 'sVkEmulation')
[clang-analyzer-core.NullDereference]
Bug: 206470603
Test: TreeHugger
Change-Id: Icaa9225afd05ec26633fa64b51047f8061defd3b
|
|
|
|
|
|
Change-Id: Iaefacf9594531b754c599b210bbbe99a8de8a4fd
|
|
|
|
Change-Id: Icf1723fc9b6593f5dd015f7cd68744c3f9ba9876
|
|
Some apps don't specify attributes.
BUG=239639751
TEST=gfxbench doesn't crash at startup
Change-Id: Ibf59952feab00a9defa1390b57c8c04a51186393
|
|
... and update handling of the created EGL surfaces and contexts.
Moves the existing Display interface to DisplaySurfaceUser as
CompositorGl needs to be bound to either the "fake window" or the
"real window" surface.
Removes TextureDraw from ContextHelper as there is only ever a
single TextureDraw instance created using the global EGL context
which is shared with every other EGL context.
Moves the existing GL ops out of PostWorker and into DisplayGl.
Upcoming changes hope to consolidate DisplayGl and DisplayVk
interface so that PostWorker does not have any knowledge of the
API specific displays.
Bug: b/233939967
Test: android build
Test: cmake build
Test: cvd start --gpu_mode=gfxstream
Test: gfxstream unit tests
Change-Id: I0d5ad35978df5b89b510e76ac7a16e59c6aef57d
|
|
... and move VkSurfaceKHR creation into DisplaySurfaceVk and
binding into DisplayVk.
Upcoming refactoring for b/233939967 would like to introduce a
DisplayGl (to parallel DisplayVk) and a Display base class in
order to fully isolate GL functionality from PostWorker. Part
of this refactoring hopes to move surface and context handling
into CompositorGl/CompositorVk and DisplayGl/DisplayVk because
handling is quite different between GL and VK components. For
example, CompositorVk doesn't actually need any surface or
context binding. This also helps hide VK specific details
inside VK specific components (DisplayVk) instead of having
them in the generic FrameBuffer.
Bug: b/233939967
Test: android build
Test: cmake build
Test: cvd start --gpu_mode=gfxstream
Test: gfxstream unit tests
Change-Id: I550b77844452941d2f7a2c8c966fe96758d257a1
|
|
|
|
This change deletes the include/EGL directory and uses the headers
from the third-party/angle path instead. It also updates the usage
of 'EGLStream' to use the 'EGLStreamKHR' type defined in the newer
headers.
Test: mkdir build && cd build && cmake .. -GNinja -DVIRGL_RENDERER_UNSTABLE_APIS=ON -DDEPENDENCY_RESOLUTION=DOWNLOAD && ninja
Change-Id: I06895a3230c2452468d2552827fb56bb358fac7f
|
|
the emulation itself.
Bug: b/241828220
Change-Id: I25fe56a737a79a8b3a47c8c87f9290f44d2daed4
|
|
Bug: 237805597
Test: presubmit
Change-Id: I747a5653bf4da29913b722fb1927bf49cabf5636
|
|
|
|
|
|
...for bad packet lengths and duplicate seqnos.
Bug: 244578445, 245275823
Test: Built and ran emulator
Change-Id: I65cc5b4347e6d85417b8d643f4a4a93b628ae401
|
|
|
|
* changes:
Move RenderContext to the new "gl" subdirectory
Move YUVConverter to the new "gl" subdirectory
Move GLESVersionDetector to the new "gl" subdirectory
|