This reverts commit dd5c7b6aeb41270f0b457c18a2a31cf136562bf0.
Reason for revert: b/366153044
Reviewed-by: Aaron Ruby <aruby@blackberry.com>
Acked-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27246>
We move the fence signaling to after color buffer copying. This is done
by reusing the existing CPU CV object and wait on it before triggering
the GPU vkWaitForFences.
Also rewrite fenceInfo.status with std::future to simplify the code.
Design: go/emu-async-cb-copy
Reviewed-by: Aaron Ruby <aruby@blackberry.com>
Acked-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27246>
The vkGetFenceStatus() call can not be sent to the host for fences
that have imported an external payload (sync fd) because the sync
fd does not exist on the host. A fence used as part of a swapchain
present may be created in the unsignaled state. Then, during
vkQueuePresentKHR() on Android, vkQueueSignalReleaseImage() is used
to import a sync fd payload into the present fence. Prior to this
change, if the user (ANGLE) does vkGetFenceStatus() on this fence,
it would never appear as signaled because the sync fd fence is not
actuallly connected to the fence on the host and the host would just
always return the VK_NOT_READY from the fence's initial unsignaled
state.
This change also updates VkFence_Info to use a std::optional<int>
to make it possible to distinguish if a fence has an imported
already-signaled payload vs not having an imported payload.
Reviewed-by: Aaron Ruby <aruby@blackberry.com>
Acked-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27246>
Removes use of android-emu/aemu/base from
libgfxsteam_vulkan.so. Also removes duplicated code
entries that are known to VK-specific. More code
can likely be deleted than currently is.
Reviewed-by: Aaron Ruby <aruby@blackberry.com>
Acked-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27246>
We probably want to recreate encoder logging, but
just delete it for now.
Reviewed-by: Aaron Ruby <aruby@blackberry.com>
Acked-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27246>
We acquire the lock when createExtraHandlecsForNextApi is called. But
depending on config, vkCreateDescriptorPool may or may not call
createExtraHandlesForNextApi, which results in inconsistency of the lock
state when calling snapshot()->vkCreateDescriptorPool.
We add a tryLock so that it always acquires the lock before modifying
mReconstruction.
Reviewed-by: Aaron Ruby <aruby@blackberry.com>
Acked-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27246>
The HealthMonitor is hang detection at the API Vulkan
encoder level.
It's actually not used in CF/AAOS/Emualtor. The
plan for AOSP is actually to increase the amount
of Android CTS that are run against gfxstream to
increase health/stability, for example (b/347288539).
So if we consistently pass CTS on main with gfxstream
(as is the plan), HealthMonitoring would be somewhat
redundant.
Also, AndroidHealthMonitor is somewhat duplicated with
libaemu's HealthMonitor as well.
Also, nuke EncoderAutoLock while we're at it. It also
is unused code.
Reviewed-by: Aaron Ruby <aruby@blackberry.com>
Acked-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27246>
Print missing structure information before calling abort.
Reviewed-by: Aaron Ruby <aruby@blackberry.com>
Acked-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27246>
... by silo'ing each physical device in its own device group.
This unblocks dEQP VK testing until the mesa layer has codegen
to fully translate structs in pNext chains.
Reviewed-by: Aaron Ruby <aruby@blackberry.com>
Acked-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27246>
- One less dependency on guest/android-emu/
- MESA_SCOPE_TRACE works on Linux guests or via Kumquat
too, for better performance analysis
Reviewed-by: Aaron Ruby <aruby@blackberry.com>
Acked-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27246>
Unboxing fails if a NULL handle is passed in, triggering an abort.
However for vkCmdBeginTransformFeedbackEXT, this is perfectly
valid:
"For each element of pCounterBuffers that is VK_NULL_HANDLE,
transform feedback will start capturing vertex data to byte zero
in the corresponding bound transform feedback buffer."
Reviewed-by: Aaron Ruby <aruby@blackberry.com>
Acked-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27246>
pCounterBuffers can be NULL, which crashes on the autogen path:
"For each element of pCounterBuffers that is VK_NULL_HANDLE,
transform feedback will start capturing vertex data to byte zero
in the corresponding bound transform feedback buffer."
Need to special case.
Intel
Reviewed-by: Aaron Ruby <aruby@blackberry.com>
Acked-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27246>
With newer versions of libstdc++, debug builds of gfxstream
hit this assert:
0x00007ffff6ed2d60 in std::__glibcxx_assert_fail
(file=<optimized out>, line=<optimized out>, function=<optimized out>,
condition=<optimized out>)
at /usr/src/debug/gcc/gcc/libstdc++-v3/src/c++11/assert_fail.cc:41
std::allocator<VkDescriptorBufferInfo> >::operator[]
(this=0x555555609380, __n=0)
at /usr/include/c++/14.1.1/bits/stl_vector.h:1130
(pDescriptorSets=0x7fffffffcc30, descriptorSetCount=2,
bufferInfos=std::vector of length 1, capacity 1 = {...})
at ../guest/vulkan/gfxstream_vk_device.cpp:718
(device=0x55555562f400, descriptorWriteCount=2,
pDescriptorWrites=0x7fffffffcc30, descriptorCopyCount=0,
pDescriptorCopies=0x0)
at ../guest/vulkan/gfxstream_vk_device.cpp:746
Use resize instead of reserve + memset.
"That way the vector size would be initialized, bounds checks would
be happy, and default-init would automatically zero out POD structs
for us." -- dextero@
Reviewed-by: Aaron Ruby <aruby@blackberry.com>
Acked-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27246>
New protocol which instructs the host to share sync descriptor
with ExternalObjectManager.
Reviewed-by: Aaron Ruby <aruby@blackberry.com>
Acked-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27246>
Extensions known in the vulkan header will be displayed with their
names to make debugging easier and the missing feature more visible.
Reviewed-by: Aaron Ruby <aruby@blackberry.com>
Acked-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27246>
Reduce clutter due to extensions related ifdef code block generation
and avoid changing many files when a new extension support is added
with partial functionality.
Reviewed-by: Aaron Ruby <aruby@blackberry.com>
Acked-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27246>
This reverts commit e43734cb0fa4e938f6ca43a01ed82d5921ad3f0d.
Reason for revert: have a real fix in aosp/3151746
Reviewed-by: Aaron Ruby <aruby@blackberry.com>
Acked-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27246>
We did a few things here.
1. Update VkBindImageMemoryInfo in vkBindImageMemory2 for snapshot. In
its implementation, vkBindImageMemory2 sometimes replaces an unboxed
VkImage handle with a new one, which breaks snapshot. We update its
value so that it can map to the proper one when calling snapshot
functions. Note that there is a const cast which we might want to fix
later.
2. Lock it properly when calling createExtraHandlesForNextApi.
3. Add support for more VkImage format. Fix mipmap extend calculation.
Now it works with -guest-angle but fails with VVL, not sure why.
Reviewed-by: Aaron Ruby <aruby@blackberry.com>
Acked-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27246>
This reverts commit 58934c4400e20934ef3172aab1a5ebefaa08e786.
Reason for revert: Breaks meson build
Same issue has aosp/3141746
Reviewed-by: Aaron Ruby <aruby@blackberry.com>
Acked-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27246>
The pattern of vkBindImageMemory2 doesn't really fit with what we have
in the auto-code generator. Thus we need to implement a special entry
for it in the code generator. In future we might want to split the code
and directly write c++ code for similar APIs.
Note that due to complexity we only support vkBindImageMemory2 with
bindInfoCount == 1, for now.
Reviewed-by: Aaron Ruby <aruby@blackberry.com>
Acked-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27246>
These APIs and the related extension are deprecated in favor of the new
metal_objects extension to make sure the loader can intercept them, so that the layers, such as VVL, would work fine.
Removes necessity to load functions directly from a separate dylib.
VK_MVK_moltenvk extension support is still passed to the guest side
due to the existing checks for enabling the AHB extension.
Reviewed-by: Aaron Ruby <aruby@blackberry.com>
Acked-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27246>
This reverts commit c4de357e4f0b256a246007982aacde2dedc10d15.
Reviewed-by: Aaron Ruby <aruby@blackberry.com>
Acked-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27246>
In our implementation, VkQueue is created in vkCreateDevice instead of
vkGetDeviceQueue. We will need to track their creation API properly,
otherwise there will be a name mismatch on snapshot load. This will
result in a crash during snapshot load with -guest-angle.
This commit marks their creation API properly.
Same thing applies for VkDescriptorSet.
Reviewed-by: Aaron Ruby <aruby@blackberry.com>
Acked-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27246>
Not required for any Mesa object backings.
TODO: Investigate why VkBuffer still required (and crashing without
translation)
Reviewed-by: Aaron Ruby <aruby@blackberry.com>
Acked-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27246>
Record all vkCmd* commands. Also clear them on vkResetCommandBuffer.
Also use VK_IMAGE_ASPECT_STENCIL_BIT when snapshotting stencil
images.
Reviewed-by: Aaron Ruby <aruby@blackberry.com>
Acked-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27246>
Original approach was to translate all possible handle types for complete
Mesa object equivalence. Practically, only a subset of these objects
actually need to be real Mesa objects for use with WSI. This simplifies
the translation and handwritten entries substantially, which is somewhat
brittle and bug-prone.
Clean up parts of functable.py that we don't need.
Reviewed-by: Aaron Ruby <aruby@blackberry.com>
Acked-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27246>
In preparation for making the issue less likely to appear.
This reverts commit 696eaaa012e369157dd745d1bf083b4dc064deaf.
Reviewed-by: Aaron Ruby <aruby@blackberry.com>
Acked-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27246>
This reverts commit 9eef6d0aefcf0aa1c07d42d9b307b1092a6deec9.
... as this does not yet have a way to generically convert Mesa
handles into Gfxstream handles in extension structs which causes
breakage in dEQP VK for many tests.
Moves the mesa based codegen to a separate `mesa_func_table`.
Reland fixes the end2end test guest vulkan ICD.
cts -m CtsDeqpTestCases
--module-arg CtsDeqpTestCases:include-filter:dEQP-VK.*
Reviewed-by: Aaron Ruby <aruby@blackberry.com>
Acked-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27246>
We have a situation that if an image memory is created with
VkMemoryDedicatedAllocateInfo, the following commands that refer to the
same image must happen in this exact order:
vkCreateImage
vkAllocateMemory (with dedicated image memory)
vkBindImageMemory
vkCreateImageView
Our previous dependency graph implementation uses Vulkan objects as
nodes and does not haven enough granularity to handle this situation.
vkCreateImageView can only be executed after an image is created and
bound to memory. Our previous dependency graph does not have the
notion of "bound to memory". We worked around it by adding
vkBindImageMemory as part of the image initialization steps.
To make sure we can bind image at initialization, we marked image to be
dependent on the memory it bounds to. When adding the new
vkAllocateMemory with dedicated image memory extension into the mix, it
introduced a circular dependency. vkAllocateMemory will say the memory
has dependency on the image. vkBindImageMemory will say the image has
dependency on the memory.
To properly resolve the issue, we will need to record the state of an
Vulkan object, and declare that vkCreateImageView not only depends on an
image, but also requires the image to be bound to memory. We do so by
introducing a "State" enum and use the pair <VkObjectHandle, State> as
the vertex in the deppendency graph. Currently State is an enum with 2
values: CREATED and BOUND_MEMORY. By default, after a VkCreate* command,
an object will be in CREATED state. When calling vkBindImageMemory or
vkBindBufferMemory it will transform the image or buffer into
BOUND_MEMORY state. Then vkCreateImageView will have dependency with
{VkImage, BOUND_MEMORY}.
Then we can create a dependency graph that looks like this:
VkImageView --> {VkImage,BOUND_MEMORY} --> VkMemory
| |
| ┌-------------------⅃
v v
VkImage
No more circular dependency.
Reviewed-by: Aaron Ruby <aruby@blackberry.com>
Acked-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27246>
This reverts commit d6e1b00029419b3ae52607967bbb950137bff848.
Reason for revert: broke end2end tests b/333885743
Reviewed-by: Aaron Ruby <aruby@blackberry.com>
Acked-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27246>
... as this does not yet have a way to generically convert Mesa
handles into Gfxstream handles in extension structs which causes
breakage in dEQP VK for many tests.
Moves the mesa based codegen to a separate `mesa_func_table`.
cts -m CtsDeqpTestCases
--module-arg CtsDeqpTestCases:include-filter:dEQP-VK.*
Reviewed-by: Aaron Ruby <aruby@blackberry.com>
Acked-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27246>
There are 2 issues with the current implementation of VkReconstruct:
a. When deleting a child handle it does not erase itself from its
parent. When doing snapshot we enumerate children handles and check
if they are still "alive". But the handle pool is written in the way
that it could reuse handles that have been previously destroyed,
invalidating the "alive" check.
b. Previous dependency graph does not support one handle with multiple
parents.
In this commit we fix (a) by making child<->parent pointer
bidirectional. When deleting a child it will be erased from its parents.
We fix (b) by rewriting a significant part of dependency graph logic on
snapshot save.
The commit also makes all destroy command to be recursive (except for
vkDestroyShaderModule), which will be handled in later commits.
Also add dependency VkFrameBuffer -> VkImageView.
Reviewed-by: Aaron Ruby <aruby@blackberry.com>
Acked-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27246>
We are trying to make vk snapshot load the google Chrome home page
without crashing. With this CL it does not crash during load but would
crash with a device lost after submitting a queue (which is
unsurprising because everything is missing to snapshot vk queue).
Content of this commit includes:
- Add dependencies for VkImageView, VkGraphicsPipelines, VkFramebuffer.
They are necessary to tell the snapshot module the order of vk object
creation.
- Add vkBindImageMemory into the loading sequencey of VkImage, so that
it would be executed before vkCreateImageView.
- Delay the destruction of VkShaderModule. This is because other
objects can still refer to it after it is destroyed.
- Initialize VK backend for color buffer properly on snapshot load.
- Save and load vk images in the same order by sorting them according
to their boxed handle.
- Record all the placeholder handles for vkCreateDescriptorPool. For
performance purpose this function creates a lot of extra handles
without real contents. We need to snapshot those handles as well.
Reviewed-by: Aaron Ruby <aruby@blackberry.com>
Acked-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27246>
... which will allow building Gfxstream from a fixed version of
AEMU for the purposes of packaging.
Reviewed-by: Aaron Ruby <aruby@blackberry.com>
Acked-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27246>
Sometimes guest renders to an AHB without calling
vkQueueSignalReleaseImageANDROIDAsyncGOOGLE. This would result in the
color buffer not being updated from Vulkan.
This commit tracks the situation that AHB is rendered to, and copies its
content to color buffer.
Note that it adds extra wait, which could hurt performance.
Reviewed-by: Aaron Ruby <aruby@blackberry.com>
Acked-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27246>
This commit snapshots vk image content by allocating a staging buffer
and copying the bytes on snapshot. It only works in the simplest setup.
Many situations are not considered in this commit, they include:
(1) the image does not support VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL
layout;
(2) the image does not support VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL
layout;
(3) the queue is dirty.
Also there is no performance optimization.
Implementation-wise, snapshot happens in VkDecoderGlobalState after
recording / playing back all create / bind commands. It borrows an
existing queue to run the extra vk copy commands. A temporary staging
buffer is also created for copying. Later we could optimize the code
by reusing most of the temporary objects.
Reviewed-by: Aaron Ruby <aruby@blackberry.com>
Acked-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27246>
This commit tracks more functions related to coherent memory snapshot
and saves the memory content.
At this point we are not sure if we need to manually copy the memory
content, because they are supposed to be cloned in RAM snapshot. We
could revert that part if they turn out to be unnecessary in future.
Reviewed-by: Aaron Ruby <aruby@blackberry.com>
Acked-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27246>
The call has to be processed using the internal descriptor logic.
Reviewed-by: Aaron Ruby <aruby@blackberry.com>
Acked-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27246>
This avoids compilation errors when codegen'ing at build
time.
Reviewed-by: Aaron Ruby <aruby@blackberry.com>
Acked-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27246>
Do the same with vkQueueCommitDescriptorSetUpdatesGOOGLE
Reviewed-by: Aaron Ruby <aruby@blackberry.com>
Acked-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27246>
This is a mega-change to support Linux guest WSI with gfxstream.
We tried to do a branch where every commit was buildable and
runnable, but that quickly proved unworkable. So we squashed
the branch into a mega-change.
Zink provides the GL implementation for Linux guests, so we just
needed to implement the proper Vulkan Wayland/X11 WSI
entrypoints.
The overall strategy to support this is to use Mesa's WSI
functions. The Vulkan WSI layer was also considered:
https://gitlab.freedesktop.org/mesa/vulkan-wsi-layer
But it was less maintained compared to Mesa. The way Mesa common
layers communicate with drivers is the through base objects
embedded in driver and a common dispatch layer:
https://gitlab.freedesktop.org/mesa/mesa/-/blob/main/docs/vulkan/dispatch.rsthttps://gitlab.freedesktop.org/mesa/mesa/-/blob/main/docs/vulkan/base-objs.rst
Our objects are defined in gfxstream_vk_private.h. Currently,
Mesa-derived Vulkan objects just serve as shim to gfxstream
Vulkan’s internal handle mapping. Long-term, we can use
Mesa-derived objects inside gfxstream guest Vulkan exclusively.
The flow is typically inside a Vulkan entrypoint is:
- VK_FROM_HANDLE(vk-object) to convert to a gfxstream_vk_obj
object
- Call ResourceTracker::func(gfxstream_vk_obj->internal) or
VkEncoder::func(gfxstream_vk_obj>internal)
- Return result
A good follow-up cleanup would be to delete older gfxstream
objects. For example, we now have struct gfxstream_vk_device
and info_VkDevice in ResourceTracker.
Most of this logic was auto-generated and included in
func_table.cpp. Some vulkan functions were too difficult to
auto-generate or required special logic, and these are included
in gfxstream_vk_device.cpp. For example, anything that needs to
setup the HostConnection requires special handling.
Android Blueprint support is added to the parts of Mesa needed
to build the Vulkan runtime. One thing to call out it's
required to build the guest/vulkan_enc and guest/vulkan files
in the same shared library now, when previously have
libvulkan_enc.so and libvulkan_ranchu.so was sufficient
[otherwise, some weak pointer logic wouldn't work].
A side effect of this is libOpenglSystem must also be a static
lib, and so should libandroid_aemu too. That conceptually makes
sense and the Meson build had been doing this all a long. We
can probably transition everything besides libGLESv1_emulation.so,
libGLESv2_emulation.so and libvulkan_ranchu.so to be static.
This requires changes in the end2end tests, because since each
HostConnection is separate and internal to it's constituent
library. Lifetimes need to be managed separately: for example
the HostConnection instance created by the end2end tests would
not be visible inside libvulkan_ranchu.so anymore. Probably the
best solution would be improve the testing facade so a
HostConnection represents one virtio-gpu context, while some
other entity represents a virtio-gpu device (client-server
would work).
vk.xml was modified, but change sent to Khronos:
https://gitlab.khronos.org/vulkan/vulkan/-/merge_requests/6325
Fuchsia builds still need to be migrated, but they already have
Fuchsia Mesa with all the build rules so that shouldn't be too
bad. Just need to copy them over the gfxstream/Mesa hybrid.
The new command for building Linux guests is:
meson amd64-build/ -Dvulkan-drivers="gfxstream" -Dgallium-drivers="" -Dvk-no-nir=true -Dopengl=false
Big shout-out to Aaron Ruby, who did most of the gnarly codegen
needed to get the function table logic to work.
* Run Weston/vkcube on Linux and automotive platform
* launch_cvd --gpu_mode=gfxstream vkcube
* launch_cvd --gpu_mod=gfxstream_guest_angle
* vkcube + 3D Mark Slingshot extreme work with guest ANGLE and
GL-VK interop
* GfxstreamEnd2EndTests
* Some select dEQP tests
Aaron Ruby (46):
gfxstream: function table: remove entry points that are hand-written.
gfxstream: function table: more changes
gfxstream: function table: scope internal_arrays to encoder
gfxstream: function table: autogenerate compoundType params
gfxstream: add handwritten EnumeratePhysicalDeviceGroup entrypoint.
gfxstream: function table: handle nested handle arrays
gfxstream: function table: adding some handwritten implementations
gfxstream: revert some unnecessary changes
gfxstream: use vk_object_zalloc/free instead of vk_zalloc/free.
gfxstream: revert most gfxstream objects to use vk_object_base
gfxstream: function table: handwritten commmand-buffers/pools
gfxstream: codegen functionality to handle special param
gfxstream: function table: random fixes
gfxstream: add vk_command_buffer_ops handlers
gfxstream: func_table.py: Codegen support for nested compound type
gfxstream: remove handwritten/add autogen entry points
gfxstream: add gfxstream_vk_device.cpp
gfxstream: query device and instance extensions early
gfxstream: func_table: explicit allocation for nested arrays/compound
types
gfxstream: goldfish_vulkan: fix commandBuffer allocation.
gfxstream: meson: Raise api_version in ICD config to 1.1.
gfxstream: function table: add more handwritten entries
gfxstream: goldfish_vulkan: update VkDescriptorSetAllocateInfo logic
gfxstream: function table: NULL check on internal_object dereference
gfxstream: function table: Remove POSTPROCESSES handling from
functable
gfxstream: mesa: Add 'gfxstream' as a -Dvulkan-drivers
gfxstream: ResourceTracker: add some allowedExtensions
gfxstream: gfxstream_vk_device: add wsi_common_entrypoints
gfxstream: Move instance handling into gfxstream_vk_device.cpp
gfxstream: ResourceTracker: Enable Linux WSI-related extensions
gfxstream: wsi: add wsi_device initialization
gfxstream: gfxstream_vk_device: use Mesa common physical device
management
gfxstream: ResourceTracker: translate mesa objects in user buffer
gfxstream: exclude VkSampler and VkDescriptorSet objects from
translation
gfxstream: Add guest-side external memory support with colorBuffers.
gfxstream: function table: Modify semaphoreList inputs to no-op
semaphores
gfxstream: function table: Allow VK_NULL_HANDLE for free/destroy APIs.
gfxstream: cereal: Add VK_EXT_depth_clip_enable as supported feature.
gfxstream: vulkan_enc: un-namespace vk_util.h and vk_struct_id.h
gfxstream: gfxstream_vk_device.cpp: Support VK_KHR_surface and
VK_*_surface
gfxstream: vulkan_enc: Add support for Mesa-only extensions.
gfxstream: ResourceTracker: Use DEVICE_TYPE_VIRTUAL_GPU always
gfxstream: platform: add dma-buf export support with dedicatedBuffer.
gfxstream: ResourceTracker: add VK_EXT_depth_clip_enable allowed
extension
gfxstream: ResourceTracker: external memory via QNX_screen_buffer
extension
gfxstream: Add VK_QNX_external_memory_screen_buffer to VulkanDispatch
Gurchetan Singh (18):
gfxstream: mesa: write Android.bp files
gfxstream: generate gfxstream_vk_entrypoints.{c, h}
gfxstream: vulkan_enc: add gfxstream_vk_private.h (objects)
gfxstream: function table: modify function table to use gfxstream_vk_*
gfxstream: compiles
gfxstream: build system improvements
gfxstream: ResourceTracker: don't crash without
VkBindImageMemorySwapchainInfoKHR
gfxstream: vk.xml: make some vkAcquireImageANDROID params optional
gfxstream_vk_device: filter out swapchain maintenance guest side
gfxstream: end2end: fixes for End2End tests
gfxstream: func_table: custom vkEnumerateInstanceLayerProperties
gfxstream: add VK_EXT_DEBUG_UTILS_EXTENSION_NAME into Mesa list
gfxstream: clang-format guest code
gfxstream: libandroid AEMU static
gfxstream: vkEnumerateInstanceVersion
gfxstream: vkCreateComputePipeLines
gfxstream: make end2end tests happy
gfxstream: delete prior vk.xml, vk_icd_gen.py
Reviewed-by: Aaron Ruby <aruby@blackberry.com>
Acked-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27246>
