|
2450b59e
|
2023-10-16T15:23:48
|
|
Vulkan: Attempt to fix the perf/mem regression of previous CL
We should get the size before destroy garbage. Otherwise the size
returned likely is "garbage". The accounting may affect other logic and
cause memory and perf regression.
Bug: b/305752495
Bug: b/305791801
Change-Id: If89bc95c11f43ff3da61fdab3ae961c0211ca92a
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4943169
Reviewed-by: Amirali Abdolrashidi <abdolrashidi@google.com>
Reviewed-by: Roman Lavrov <romanl@google.com>
Commit-Queue: Charlie Lao <cclao@google.com>
|
|
|
90dd58a2
|
2023-10-04T11:54:22
|
|
Vulkan: Reduce mGarbageMutex lock contention
Right now mGarbageMutex is used to control access to mSharedGarbage,
mPendingSubmissionGarbage, mSuballocationGarbage,
mPendingSubmissionSuballocationGarbage and mOrphanedBufferBlocks. Some
times garbage clean up does take a bit longer time, especially on some
VM platforms. This some times causes lock contention between main render
thread and background garbage clean up thread, which defeats the benefit
of having garbage clean up in the background thread. This CL utilizes
angle::FixedQueue for garbage list so that enqueue and dequeue can be
concurrent, which avoids this lock contention.
Bug: b/302739073
Change-Id: I44b2b0e7f9f5ef438266fa277b24a2cb1606e689
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4899299
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Hailin Zhang <hailinzhang@google.com>
Commit-Queue: Charlie Lao <cclao@google.com>
|
|
|
2608c622
|
2023-10-06T13:32:49
|
|
Vulkan: Refactor SharedGarbageList into templated class
This CL mostly involves non-functional changes to prepare for next CL.
No behavior change is expected.
This CL wraps the garbage list into its own templated class which
maintains std::queue and tracks number bytes in the queue etc.
This CL also renames SharedBufferSuballocationGarbageList to
BufferSuballocationGarbageList to reduce verbosity a bit.
This CL deleted GarbageAndQueueSerial and GarbageQueue since they are no
longer being used.
This renames vk::GarbageList to vk::GarbageObjects to reduce name
confusion with SharedGarbageList.
Bug: b/302739073
Change-Id: I7370c147847ffe69ad8aa3b48251d8b5762f97f9
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4919816
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Hailin Zhang <hailinzhang@google.com>
Commit-Queue: Charlie Lao <cclao@google.com>
|
|
cd901cdd
|
2023-03-03T17:04:27
|
|
Vulkan: Add and use rx::vk::ReleasableResource class.
This is a follow up for CL:
Vulkan: Enforce ContextPriority in ShareGroup and with EGLImage
Bug: angleproject:8039
Change-Id: I9e654557d4a1ce9aee4b04f1211eeb6ae3f0e482
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4306721
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Charlie Lao <cclao@google.com>
Commit-Queue: Igor Nazarov <i.nazarov@samsung.com>
|
|
|
c402ea1c
|
2023-02-15T12:01:38
|
|
Vulkan: Rename hasUnfinishedUse to hasResourceUseFinished
Most usage of hasUnfinishedUse is for !hasUnfinishedUse, and there was
feedback that negative API is not preferred. This CL changes it to
positive API name. Similarly renamed hasUnsubmittedUse to
hasResourceUseSubmitted.
Bug: b/267348918
Change-Id: Idb10b0f998ec50116ffb6aada19a98a516e87824
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4257105
Commit-Queue: Charlie Lao <cclao@google.com>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
|
|
|
3307e8d2
|
2023-02-16T10:26:20
|
|
Vulkan: Add ostream operator for ResourceUse/QueueSerial
This is some debug logging code I have been using in the past. Want to
land it for convenience.
Bug: None
Change-Id: Ife044dd7dc733d3d38283949bdfc11b203cbe429
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4177488
Reviewed-by: Yuxin Hu <yuxinhu@google.com>
Commit-Queue: Charlie Lao <cclao@google.com>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
|
|
|
376d309c
|
2022-12-15T09:43:00
|
|
Vulkan: Remove unnecessary usesBuffer() check
There are places that we call setQueueSerial after usesBuffer() check.
This was useful when we had the ResourceList where it is more expensive
to set serial. But now setQueueSerial is cheap (actually is cheaper than
usesBuffer check), so there is no need to do this check any more. This
CL removes the check to further reduce the CPU overhead. Because of
this, mUsedBufferCount will not be accurate, so this CL also removes the
tracking of mUsedBufferCount (was only for informational purpose
anyway).
This CL also removes commandBufferQueueSerial.valid() check in
Resource::usedByCommandBuffer() and turns it into assertion. Some places
in contextVk will ensure we only call it on started renderpass so that
other places that calls usedByCommandBuffer will not need to eat the if
check.
Bug: b/262047600
Change-Id: I6b8004b6aa5b567fa94c0eb56801054f818838b1
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4112145
Reviewed-by: Yuxin Hu <yuxinhu@google.com>
Commit-Queue: Charlie Lao <cclao@google.com>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
|
|
|
cd367796
|
2022-12-12T15:10:09
|
|
Vulkan: Add assert to ensure never setQueueSerial backwards
This CL add an assertion in ResourceUse::SetQueueSerial to ensure that
we never set a serial smaller than what it already has. If that happens,
we could potentially destroy it while GPU still accessing it. With this
assertion, it exposed a bug that when a buffer is read accessed by a
renderpassCommands and then read accessed by outsideRenderPassCommands,
we were incorrectly setting the queueSerial with outsideRP's serial,
overwriting the queueSerial already set by renderPassCommands. To fix
this, this CL detects this case and keeps the queueSerial set by
renderPassCommands.
Bug: b/262047600
Change-Id: I51b17ab4a93bccd0d0b079784af96cef9d79f16f
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4099804
Reviewed-by: Yuxin Hu <yuxinhu@google.com>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Commit-Queue: Charlie Lao <cclao@google.com>
|
|
|
a703eea4
|
2022-12-09T15:55:03
|
|
Vulkan: Remove Serial::valid() check
QueueSerial right now checks if serial is valid or not in various
comparison. This CL removes valid() method from class Serial and rely on
the fact that default constructor always constructs kZeroSerial and
kZeroSerial is always smaller than any serial, i.e, always appears as
flushed, submitted and completed. This removes one branch from critical
code path where we try to detect if a buffer/image is used by a
renderpass or not etc.
Bug: b/262047600
Change-Id: Ic76fe1409d9911dc7eb86107c9a930d8bb5eaa05
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4089848
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Amirali Abdolrashidi <abdolrashidi@google.com>
Commit-Queue: Charlie Lao <cclao@google.com>
|
|
|
89cd8583
|
2022-12-12T11:23:59
|
|
Vulkan: Clean up Resource class
Resource::retainCommands() API name no longer make sense. This CL
removes retainCommands and retainReadOnly and retainReadWrite APIs and
replaced with setQueueSerial and setWriteQueueSerial call directly.
This CL also merges some of single inline functions to minimize the
file, sine the class is small anyway.
Bug: b/262048658
Change-Id: I9d16b82c79b27f3285311393601705a4ee7f6d8a
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4098005
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Commit-Queue: Charlie Lao <cclao@google.com>
Reviewed-by: Amirali Abdolrashidi <abdolrashidi@google.com>
|
|
|
f3ebb2ca
|
2022-12-08T17:02:12
|
|
Vulkan: Better mUse tracking for DynamicallyGrowingPool<Pool>
This is used only from DynamicallyGrowingPool<Pool>::onEntryFreed to set
the its tracking QueueSerial. There is better way to do this now with
per context serial. We can simply merge the QueryHelper's mUse into the
pool instead of setting pool's use to the current queueSerial. The
benefit of doing that is to possibly allow pool gets reuse/freed earlier
(i.e., more accurate tracking). This CL switches it to more accurate
tracking and removes mCurrentSerial from Context.
This CL also removes unused DynamicSemaphorePool class.
Bug: b/262054987
Change-Id: Iac3e2495cc0e3623ba63e9da7f32ad6e9c223467
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4089847
Reviewed-by: Yuxin Hu <yuxinhu@google.com>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Commit-Queue: Charlie Lao <cclao@google.com>
|
|
|
f8980c98
|
2022-12-08T11:51:42
|
|
Vulkan: Make ReadWriteResource subclass from Resource
To simplify the code slightly. The main reason for this is that the two
class variables of ReadWriteResource totally does not make sense:
mReadOnlyUse actually manes any usage, read or write. mReadWriteUse
actually means write usage.
Since Resource class's mUse means any access, subclass ReadWriteResource
from Resource class makes more sense since mUse means exact same thing
in both classes.
This CL also changes ReadQriteResource::mReadWriteUse to mWriteUse.
Bug: b/262048658
Change-Id: I0e3172a70b8cb6a6481045c46690b69fbfe9523c
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4089983
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Yuxin Hu <yuxinhu@google.com>
Commit-Queue: Charlie Lao <cclao@google.com>
|
|
|
1219f55a
|
2022-12-07T16:19:37
|
|
Vulkan: Remove Resource::isCurrentlyInUse
Due to header file include order, this function can not directly
made inline. This CL removes the function and replace it with
renderer->getUnfinishedUse() to reduce one extra function call of one
line function.
Bug: b/262048658
Change-Id: Ied33b63d0ec88336a5ce42cf7726f16b2b883b86
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4089623
Reviewed-by: Yuxin Hu <yuxinhu@google.com>
Commit-Queue: Charlie Lao <cclao@google.com>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
|
|
|
798b97b8
|
2022-12-07T15:37:51
|
|
Vulkan: mapRangeImpl should call flushImpl if unflushed write
BufferVk::mapRangeImpl() want to ensure any GPU write command has been
flushed and finished. Right now it calls flushImpl if there is any
unflushed access. It should only need to flush if there is any unflushed
*write* command. This CL changes check of any access to any write
access.
This CL also inlines isCurrentlyInUseForWrite/finishGPUWriteCommands and
removed these two single line function calls.
Bug: b/261772793
Change-Id: I1628ec31eaceb87f82e654cb1f317570ff2f6c12
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4086972
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Yuxin Hu <yuxinhu@google.com>
Commit-Queue: Charlie Lao <cclao@google.com>
|
|
|
ca263d4a
|
2022-11-23T12:11:49
|
|
Vulkan: Remove UsedInRunningCommands
The only difference of UsedInRunningCommands() and hasUnfinishedUse() is
the latter excludes the case where it has not submitted yet. But all
usages are already checked if it has been submitted or not. This CL
removes UsedInRunningCommands and uses hasUnfinishedUse instead.
Bug: b/255414841
Change-Id: I94ef0b63a0c888219cffcdfcecfa0a095fee616b
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4053262
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Yuxin Hu <yuxinhu@google.com>
Commit-Queue: Charlie Lao <cclao@google.com>
|
|
|
2e5ca217
|
2022-11-18T10:44:49
|
|
Vulkan: Let each current context has its own QueueSerial.
This CL makes every current context has its own queueSerial. At context
creation time or when context becomes current, it allocates a QueueIndex
from renderer. When it becomes non-current, it releases QueueIndex for
others to reuse. This way we significantly reduces the max number of
QueueIndexs for reasonable usage. Each CommandBuffer has its own unique
QueueSerial and we use that to determine if a resource is being used by
the given CommandBuffer. The QueueSerial for RenderPassCommands is
deferred until renderPass starts, and when we generate queueSerial for
renderPassCommands, we also reserve a range of serials for
outsideRenderPassCommands so that we can do incremental submission of
outsideRenderPassCommands without need to close renderPassCommands. In
rare situation, if that reserved serials runs out, we also close
renderPassCommands to ensure the ordering of serials matches ordering of
command buffers.
With per current context queue serial, this CL is able to set resource
queue serial as it is being used. This CL completely removes usage of
ResourceUseList class since it was introduced due to deferred setSerial.
This CL also get rid of refCount from ResourceUse since there we no
longer add it to a ResourceUseList. With that, we also able to remove
SharedResourceUse class since access to ResourceUse itself is now thread
safe since we are able to make a copy of it when we add it to
GarbageList.
Because RenderPassCommands now has its own unique QueueSerial as it
encodes command, we can use it to detect if a resource is being used by
it or not, thus this CL also removes usage of CommandBufferID.
Bug: b/255414841
Change-Id: I36dcbeaa7bc996f04e6c04bf9ad44cd0d630f61a
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4038096
Reviewed-by: Yuxin Hu <yuxinhu@google.com>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Commit-Queue: Charlie Lao <cclao@google.com>
|
|
|
7dd8478e
|
2022-11-17T10:11:02
|
|
Vulkan: Make ResourceUse::serial an FastVector of Serials
In preparation for per context queue serial, this CL makes
ResourceUse::serial a FastVector of Serial. Right now we still limited
to one serial index so that it still work the same way as before. This
CL adds necessary data type and change the function names to reflect
that tracking GPU progress needs a ResourceUse object instead of a
single Serial number.
Bug: b/255414841
Change-Id: Ic60cdf5ec8da45d1821f65a55947f5c553f65737
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4034548
Reviewed-by: Yuxin Hu <yuxinhu@google.com>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Commit-Queue: Charlie Lao <cclao@google.com>
|
|
|
8378032e
|
2022-11-15T11:47:55
|
|
Vulkan: Remove get API for mLastCompletedQueueSerial
In preparation for per context queue serial, this CL makes
mLastSubmittedQueueSerial and mLastCompletedQueueSerial private to
CommandQueue. Before this CL, we have a get function to return the last
submitted serial and last completed serial and passing these serials
around. This works because the serial is a single uint64_t number. With
per context queue serial, this will be an array of serials and there is
potential risk associated with access it from different threads. This CL
makes these serials private to CommandQueue and when you want to know if
GPU is completed with resource, you ask RendererVk/CommandQueue
directly. This way we can ensure they have thread safe access in the
CommandQueue (no lock is necessary, but all access will be restricted to
one class).
Bug: b/255414841
Change-Id: Ica565decce4a80588e0b447e179a2b634b55d7c3
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4021676
Reviewed-by: Yuxin Hu <yuxinhu@google.com>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Commit-Queue: Charlie Lao <cclao@google.com>
|
|
|
8df0c53e
|
2022-07-11T16:56:26
|
|
Vulkan: Split suballocation out into it's own file
Right now Suballocation and BufferBlock classes are in vk_utils.h, which
is included by other header files. This made it difficult to have these
two classes use variables defined in other headers. And vk_utils.h is
really designed for utilities, it isn't the best place for
bufferBlock/suballocation by definition. This CL split out suballocation
code into its own file to make future CLs much easier. No functional
change is expected other than move the code around.
Bug: b/237686097
Change-Id: I36733468b4bed152a19d9c9bca8e7459c11c3b43
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/3756761
Reviewed-by: Yuxin Hu <yuxinhu@google.com>
Commit-Queue: Charlie Lao <cclao@google.com>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
|
|
1ce69722
|
2022-05-18T13:05:22
|
|
Vulkan: Track used command buffers in ResourceUse.
The list of command buffers to ResourceUse will replace tracking
resources in a CommandBufferHelper. Currently the two tracking
methods live side-by-side, and the old method will be removed in
a future CL.
Bug: angleproject:5664
Change-Id: Ia04d77e72c508e10b549db8c8dd5f0472e4edc83
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/3656069
Reviewed-by: Charlie Lao <cclao@google.com>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
|
51dc3c3e
|
2022-05-02T18:11:26
|
|
Vulkan: Force prune when there are lots of suballocations destroyed
If there are a lot of suballocations destroyed, there is bigger chance
that some buffers become empty and we might able to trim down excessive
empty buffers. This CL tracks the suballocations destroyed at each
cleanupGarbage call and use that information to force immediate
pruneEmptyBuffers call.
Bug: b/230538246
Change-Id: Icca2ea731639545c635c09d58a8606d67405e1a6
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/3620981
Reviewed-by: Amirali Abdolrashidi <abdolrashidi@google.com>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Commit-Queue: Charlie Lao <cclao@google.com>
|
|
3587e217
|
2022-04-13T16:52:38
|
|
Remove copyResourceUseList from ShareGroupVk
* Removed copyResourceUseList() and the copy() functions in them,
as they are no longer used in submitting commands.
Bug: angleproject:7103
Change-Id: Ic62b21817aa2843f90695a8f50b79d254ec89929
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/3587531
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Commit-Queue: Amirali Abdolrashidi <abdolrashidi@google.com>
|
|
|
fe28a429
|
2022-03-30T15:34:49
|
|
Vulkan: Create buffer for vertex array if robust enabled
If robust access is enabled (i.e., chrome), we want to ensure vulkan
driver never access beyond that OpenGL buffer boundary. But with
suballocation from BufferPool, we are using the same VkBuffer for all
suballocations from the same BufferBlock. this combined with the fact
that there is no size information in the vkCmdBindVertexBuffers, it
means vulkan driver can not properly ensure vertex access not go beyond
the subrange. It can only guarantee not access beyond the entire
VkBuffer size. This CL creates a dedicated vkBuffer object and bind it
to the suballocation of the vkDeviceMemory so that vulkan driver will
see the exact range of the subrange instead of entire buffer. Since we
may allocated more memory than actual requested size and the extra
paddings are not zero filled , user size is used to create this
vkBuffer. This is only enabled when robust access is enabled.
This CL also ported webgl conformance test
out-of-bounds-index-buffers.html and out-of-bounds-array-buffers.html
to end2end test.
Bug: chromium:1310038
Change-Id: I3499ae600028149b1039082e5011232b3e4e5e80
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/3553940
Reviewed-by: Yuxin Hu <yuxinhu@google.com>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Commit-Queue: Charlie Lao <cclao@google.com>
|
|
|
730c1271
|
2022-01-07T13:41:08
|
|
Vulkan: Submit queue more often for texture data
Outside command buffers should be flushed more often in order
to prevent the texture data accumulation just before the first
render pass when they are referenced.
* Added a tracker next to copyBufferToImage() for texture size
(in ContextVk). When its value passes kMaxBufferToImageCopySize,
the outside command buffer operations should be submitted and
the tracker would be reset. Currently, the threshold value is
set to 1 << 28 = 256M.
* Added a variation of submitFrame() to be used in outside
command buffer submission. The main difference is that it
copies mResourceUseList into GetShareGroupVk() rather than
move it.
* Refactored the two functions into submitFrameImpl().
* Added a helper function to submit the outside command
buffer.
* Added explicit copy functions for ResourceUseList and
SharedResourceUse. The counter in the copied object is
incremented by 1.
* Added a test to make sure submitting the outside command
buffer does not break the render pass.
Bug: angleproject:6354
Change-Id: Ia1d4f857fcbd06934609c94622ccbf675b3b1c72
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/3379231
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Charlie Lao <cclao@google.com>
Commit-Queue: Amirali Abdolrashidi <abdolrashidi@google.com>
|
|
|
b72718d2
|
2022-02-22T16:18:45
|
|
Vulkan: Fix the data race for mUse from two threads
A data race of mUse object between two threads can occur with the
following sequences:
1) You use the buffer in the context, it adds into mResourceUseList
2) You release the buffer. Now it goes into mGarbageList, even though it
is still in the mResourceUseList.
Now this mUse object has references from two lists, the mGarbageList and
mResourceUseList, and they do not use the same mutex lock. This means
the race could happen and corrupt the mUse data. The same thing could
happen with ImageHelper object as well. This quick fix also grabs
mGrabageMutex while processing mResourceUseList. This CL creates a new
garbage list to hold garbage that has not been submitted to vulkan. And
this list will only accessed from submission thread and with
mGarbageMutex lock held. The advantage of this is that mSharedGarbage
will only have objects that already submitted, which means it is in FIFO
order so that we can break out the loop as soon as we see an uncompleted
garbage.
This bug was exposed by
MultithreadingTest.MultiContextClear/ES3_Vulkan_SwiftShader on
linux-tsan-test
Bug: angleproject:7045
Change-Id: I264c970579aaa53373a61ff067fa0e21eb410ae6
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/3482158
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Ian Elliott <ianelliott@google.com>
Commit-Queue: Charlie Lao <cclao@google.com>
|
|
|
bc3be5a8
|
2022-01-27T12:12:58
|
|
Vulkan: Add a dedicated suballocation garbage list
Suballocations are the most common garbage objects in most usage cases.
The current garbage collection code will construct a garbage object from
suballocation and then construct a SharedGarbage object with a
std::vector that holds only one element. And then it adds this
SharedGarbage to the garbage list. This CL tries to avoid create
std::vector with just one element and avoid the cost of switch statement
for each garbage object by adding a new dedicated garbage list that only
holds the suballocation garbages, which is the most common garbages in
the system. With gardenscapes running offscreen with --minimum-gpu-work,
it reduces CPU overhead from 2.55ms to 2.20ms on Pixel6.
Bug: b/215768827
Change-Id: Ia2872442462917c0caadb263769a1cbf3dd7366f
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/3414356
Reviewed-by: Tim Van Patten <timvp@google.com>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Commit-Queue: Charlie Lao <cclao@google.com>
|
|
|
a7970889
|
2022-01-24T11:50:27
|
|
Vulkan: Use queue instead of vector for mSharedGarbageList
The garbage collector code is iterating the mSharedGarbageList from
begin to end and if the garbage object is completed, we destroy it and
erase it from the vector. This is done one by one. The problem with that
is each erase call end up moving the remaining vector to ahead by one,
and this process is repeated for each completed element. This is O(N^2)
problem and really magnifies when you have many garbage objects. Since
we only add garbage when it is GPU pending, so most of these garbages
are in sort of fifo order anyway. This CL changes the garbage list from
vector to queue so that erase becomes pop and this becomes a O(N)
problem. This reduces CPU overhead as measured with minimize-gpu-work
for gardenscapes from 10ms to 2ms due to suballocation generates lots of
garbages. But this CL is a general improvement for the garbage
collection performance.
Bug: b/215768827
Change-Id: I002aadf75f958d8b79eb6281a2608597776e908d
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/3414354
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Commit-Queue: Charlie Lao <cclao@google.com>
|
|
|
0e49a3dd
|
2022-01-04T11:23:54
|
|
Vulkan: Add std::move support for BufferHelper
There are needs to support std::move for BufferHelpers in other CLs (See
crrev.com/c/3352489). Without this support, we can not store
BufferHelper into std::vector. This CL adds move support for
BufferHelper class.
Bug: b/208323792
Change-Id: I93f79490715750abc1bcedd41b683ad0c2460ebb
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/3366855
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Commit-Queue: Charlie Lao <cclao@google.com>
|
|
8f6f5a4b
|
2021-11-30T23:48:30
|
|
Vulkan: Fix image respecify's usage tracking
When respecifying an image due to mip level count changes, the previous
image is staged as an update to the new image. The resource usage info
was not being transferred to the image being staged as an update,
causing it to be prematurely deleted.
Test based on one authored by sugoi@google.com.
Bug: chromium:1270658
Bug: angleproject:4835
Change-Id: I215c65ba700d7be608d0910d3cb37fcfdf297a2a
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/3308921
Commit-Queue: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
|
|
|
dbc0c646
|
2021-11-06T01:09:26
|
|
Vulkan: Output the reason for RP closure in command buffer
To make it easier when viewing the command buffer in a graphics
debugger, this change inserts a marker just before closing the render
pass that specifies why the render pass was closed.
Bug: angleproject:2472
Change-Id: I862e500cd58332d6e199c853315c560fe6a73dc2
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/3265609
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Tim Van Patten <timvp@google.com>
Commit-Queue: Shahbaz Youssefi <syoussefi@chromium.org>
|
|
|
bae19e06
|
2021-10-26T13:35:57
|
|
Vulkan: Avoid unnecessary wait if mapBufferRange indicates read only
When we call BufferVk::mapRangeImpl(), both from internal code paths for
data reads or due to glMapBufferRange call, we are not passing the
access bit to the call. This CL passes the proper access bits to the
call and only wait for GPU writes to finish if access is for read only.
This CL also adds access bitfield to the BufferVk::mapImpl() API and
have various callers pass in the proper access bits as well.
Bug: b/203582620
Change-Id: Ica8493c902dbd7b15996266c81ce0fd4dbfc2520
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/3245487
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Tim Van Patten <timvp@google.com>
Commit-Queue: Charlie Lao <cclao@google.com>
|
|
57d59e83
|
2021-09-07T17:41:11
|
|
Vulkan: Add ResourceWrite to track Read and Write Access
vk::Resource currently only tracks accesses in general, not which type
of access is being performed. This CL adds the new class ResourceWrite
to track whether the access is a Read or Read/Write access and when
the access completes.
This allows a follow-on CL to know when a buffer is being written to by
the GPU or if the GPU is only reading from a buffer. Tracking write
accesses to buffers is required when attempting to "Ghost" (duplicate)
GPU-read-only buffers to prevent breaking the render pass when the CPU
maps the buffer memory.
Bug: angleproject:5971
Test: ComputeShaderTest.ImageBufferMapWrite
Change-Id: I965e3e75730719ccce77334744ae4feae33c6101
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/3146319
Commit-Queue: Tim Van Patten <timvp@google.com>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Charlie Lao <cclao@google.com>
|
|
|
909ea88b
|
2020-11-20T13:07:53
|
|
Reland "Vulkan: Ignore glFlush to reduce vkQueueSubmits in Asphalt 9"
This is a reland of 5cf7472dd161bbda329dfc5e4e65bb6ce0c06fbd
The ShareGroupVk::mResourceUseLists was not being cleared each call to
RendererVk::submitFrame(), so it was growing indefinitely. Each
vk::ResourceUseList within it was cleared, so it was holding an
essentially "infinite" list of empty lists, but that caused the loop in
RendererVk::submitFrame() to take more and more time until the tests
timed out.
The fix is to do 'resourceUseLists.clear()' once the loop to release all
resources has completed, like releaseResourceUsesAndUpdateSerials() does
for each individual list. Additionally, ASSERTs are added to guarantee
that the lists are empty when the ContextVk and ShareGroupVk are
destroyed.
Original change's description:
> Vulkan: Ignore glFlush to reduce vkQueueSubmits in Asphalt 9
>
> Multithreaded apps can use the following pattern:
>
> glDrawElements()
> glFenceSync()
> glFlush()
> glWaitSync()
>
> This currently results in a vkQueueSubmit for every glFlush() to ensure
> that the work has landed in the command queue in the correct order.
> However, ANGLE can instead avoid the vkQueueSubmit during the glFlush()
> in this situation by instead flushing the ContextVk's commands and
> ending the render pass to ensure the commands are submitted in the
> correct order to the renderer. This improves performance for Asphalt 9
> by reducing frame times from 150-200msec to 35-55msec.
>
> Specifically, ANGLE will call flushCommandsAndEndRenderPass() when
> there is a sync object pending a flush or if the ContextVk is currently
> shared.
>
> Additionally, on all devices except Qualcomm, ANGLE can ignore all other
> glFlush() calls entirely and return immediately. For Qualcomm devices,
> ANGLE is still required to perform a full flush (resulting in a
> vkQueueSubmit), since ignoring the glFlush() reduces the Manhattan 3.0
> offscreen score by ~3%.
>
> Bug: angleproject:5306
> Bug: angleproject:5425
> Change-Id: I9d747caf5bf306166be0fec630a78caf41208c27
> Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2552718
> Commit-Queue: Tim Van Patten <timvp@google.com>
> Reviewed-by: Charlie Lao <cclao@google.com>
> Reviewed-by: Jamie Madill <jmadill@chromium.org>
Bug: angleproject:5306
Bug: angleproject:5425
Bug: angleproject:5470
Change-Id: I14ee424d032f22e5285d67accbec078ad1955dd0
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2595811
Commit-Queue: Tim Van Patten <timvp@google.com>
Reviewed-by: Charlie Lao <cclao@google.com>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
|
|
|
60641abc
|
2020-11-16T16:12:31
|
|
Vulkan: Make ImageViewHelper a Resource
Bug: angleproject:3573
Change-Id: I12e70418b3b971e802bc911409e170cbf8c61915
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2542223
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Tim Van Patten <timvp@google.com>
Commit-Queue: Shahbaz Youssefi <syoussefi@chromium.org>
|
|
|
fcd35965
|
2020-09-29T14:15:51
|
|
Vulkan: Make DescriptorPoolHelper a Resource
Descriptor pools need to live as long as the descriptor sets that are
allocated from them. Using Serials while building a command to judge a
pool's lifetime is prone to errors, since a command's Serial value
isn't known until the command is submitted, leading to deleting pools
too early relative to when the descriptor set is actually used.
This CL updates DescriptorPoolHelper to inherit from Resource, so the
descriptor pools can be retain()'ed. This allows the Resource's counter
to indicate that a pool is in use until the command's Serial is known
and can be recorded to indicate when the command completes. This
prevents descriptor pools from being destroyed before the command
completes (while the descriptor sets are still in use), or even before
the command has been submitted. Destroying a descriptor pool resets all
of the descriptors that were allocated from it, which can trigger a
variety of VVL errors depending on when it's erroneously performed.
This CL also adds the necessary retain() calls for the descriptor pools.
In particular, the pools need to be retained each time a cached
descriptor set that was allocated from it is re-used. This is relatively
simple with the current design, since we always clear the descriptor set
caches whenever a new pool is allocated, so the descriptor pool binding
is always accurate.
Bug: angleproject:5030
Test: VulkanMultithreadingTest::MultiContextDrawSmallDescriptorPools()
Change-Id: Iac9e7efef338f169a6bf8ac3b2140e03dd326641
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2504457
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Courtney Goeltzenleuchter <courtneygo@google.com>
Commit-Queue: Tim Van Patten <timvp@google.com>
|
|
|
9e3eec54
|
2020-10-25T15:44:09
|
|
Revert "Vulkan: Make DescriptorPoolHelper a Resource"
This reverts commit 5dcd29a6e532e4bd617af8767d488120b57f3b2c.
Reason for revert: Breaking the ANGLE -> Chromium roller:
https://chromium-review.googlesource.com/c/chromium/src/+/2496281
Original change's description:
> Vulkan: Make DescriptorPoolHelper a Resource
>
> Descriptor pools need to live as long as the descriptor sets that are
> allocated from them. Using Serials while building a command to judge a
> pool's lifetime is prone to errors, since a command's Serial value
> isn't known until the command is submitted, leading to deleting pools
> too early relative to when the descriptor set is actually used.
>
> This CL updates DescriptorPoolHelper to inherit from Resource, so the
> descriptor pools can be retain()'ed. This allows the Resource's counter
> to indicate that a pool is in use until the command's Serial is known
> and can be recorded to indicate when the command completes. This
> prevents descriptor pools from being destroyed before the command
> completes (while the descriptor sets are still in use), or even before
> the command has been submitted. Destroying a descriptor pool resets all
> of the descriptors that were allocated from it, which can trigger a
> variety of VVL errors depending on when it's erroneously performed.
>
> This CL also adds the necessary retain() calls for the descriptor pools.
> In particular, the pools need to be retained each time a cached
> descriptor set that was allocated from it is re-used. This is relatively
> simple with the current design, since we always clear the descriptor set
> caches whenever a new pool is allocated, so the descriptor pool binding
> is always accurate.
>
> Bug: angleproject:5030
> Test: MultithreadingTest::MultiContextDrawSmallDescriptorPools()
> Change-Id: I5fdeeb46159448dfd679d7169e423048348be5ab
> Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2437609
> Commit-Queue: Tim Van Patten <timvp@google.com>
> Reviewed-by: Jamie Madill <jmadill@chromium.org>
> Reviewed-by: Courtney Goeltzenleuchter <courtneygo@google.com>
TBR=courtneygo@google.com,timvp@google.com,jmadill@chromium.org
# Not skipping CQ checks because original CL landed > 1 day ago.
Bug: angleproject:5030
Change-Id: I0fd6d9a0e1b0989b22368ef98652281288699deb
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2497222
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
|
|
|
5dcd29a6
|
2020-09-29T14:15:51
|
|
Vulkan: Make DescriptorPoolHelper a Resource
Descriptor pools need to live as long as the descriptor sets that are
allocated from them. Using Serials while building a command to judge a
pool's lifetime is prone to errors, since a command's Serial value
isn't known until the command is submitted, leading to deleting pools
too early relative to when the descriptor set is actually used.
This CL updates DescriptorPoolHelper to inherit from Resource, so the
descriptor pools can be retain()'ed. This allows the Resource's counter
to indicate that a pool is in use until the command's Serial is known
and can be recorded to indicate when the command completes. This
prevents descriptor pools from being destroyed before the command
completes (while the descriptor sets are still in use), or even before
the command has been submitted. Destroying a descriptor pool resets all
of the descriptors that were allocated from it, which can trigger a
variety of VVL errors depending on when it's erroneously performed.
This CL also adds the necessary retain() calls for the descriptor pools.
In particular, the pools need to be retained each time a cached
descriptor set that was allocated from it is re-used. This is relatively
simple with the current design, since we always clear the descriptor set
caches whenever a new pool is allocated, so the descriptor pool binding
is always accurate.
Bug: angleproject:5030
Test: MultithreadingTest::MultiContextDrawSmallDescriptorPools()
Change-Id: I5fdeeb46159448dfd679d7169e423048348be5ab
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2437609
Commit-Queue: Tim Van Patten <timvp@google.com>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Courtney Goeltzenleuchter <courtneygo@google.com>
|
|
ed876984
|
2020-10-03T11:00:36
|
|
Vulkan: functionally complete worker thread
Working on enhancing worker thread to completely own primary command
buffers. This will include not only processing SCBs from main thread
into a primary, but also submitting those command buffers to the queue.
The CommandProcessor is a vk::Context so it can handle errors in the
worker thread. When the main thread submits tasks to the worker
thread it also syncs any outstanding errors from the worker.
Include asynchronousCommandProcessing feature that will control
whether the worker thread task does it's work in parallel or not. If
false, we wait for the thread to complete it's work before letting the
main thread continue. If true, the thread can execute in parallel with
the main thread.
Bug: b/154030730
Bug: b/161912801
Change-Id: I00f8f013d6cbb2af12a172c4f7927855db2f0ebf
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2328992
Commit-Queue: Courtney Goeltzenleuchter <courtneygo@google.com>
Reviewed-by: Tim Van Patten <timvp@google.com>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
|
|
|
68a5baeb
|
2020-09-23T22:13:03
|
|
Revert "Vulkan: Implement a SharedResourceUse pool"
This reverts commit de335c16855f11d1f0a6f0b37bee30c8a09a6c1d.
Reason for revert: Might actually regress CPU overhead perf.
Unsure but it's possible the reported perf improvement was due
to variance.
Original change's description:
> Vulkan: Implement a SharedResourceUse pool
>
> When adding a Resource to the ResourceUseList of ContextVk
> we constructed a new SharedResourceUse object for tracking
> and update of the Resource's Serial. We would then delete
> it after releasing the resource. This incurs repeated
> memory operation costs.
>
> Instead we now allocate a pool of SharedResourceUse objects
> and acquire and release from this pool as needed.
>
> VTune profile of the Manhattan 30 offscreen benchmark
> shows the CPU occupancy of bufferRead decrease from an
> average of 0.9% -> 0.6% and imageRead decreases from
> an average of 0.4% -> 0.3%. The bottleneck for both
> these methods is the retain() method that leverages
> the new SharedResourceUse pool.
>
> Bug: angleproject:4950
> Change-Id: Ib4f67c6f101d4b2de118014546e6cc14ad108703
> Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2396597
> Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
> Reviewed-by: Jamie Madill <jmadill@chromium.org>
> Commit-Queue: Mohan Maiya <m.maiya@samsung.com>
TBR=syoussefi@chromium.org,jmadill@chromium.org,m.maiya@samsung.com
# Not skipping CQ checks because original CL landed > 1 day ago.
Bug: angleproject:4950
Change-Id: I40081551c3db67d6e55182fea40119946ed16ac3
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2426479
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
de335c16
|
2020-09-14T12:04:20
|
|
Vulkan: Implement a SharedResourceUse pool
When adding a Resource to the ResourceUseList of ContextVk
we constructed a new SharedResourceUse object for tracking
and update of the Resource's Serial. We would then delete
it after releasing the resource. This incurs repeated
memory operation costs.
Instead we now allocate a pool of SharedResourceUse objects
and acquire and release from this pool as needed.
VTune profile of the Manhattan 30 offscreen benchmark
shows the CPU occupancy of bufferRead decrease from an
average of 0.9% -> 0.6% and imageRead decreases from
an average of 0.4% -> 0.3%. The bottleneck for both
these methods is the retain() method that leverages
the new SharedResourceUse pool.
Bug: angleproject:4950
Change-Id: Ib4f67c6f101d4b2de118014546e6cc14ad108703
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2396597
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Commit-Queue: Mohan Maiya <m.maiya@samsung.com>
|
|
|
295d2ccd
|
2020-08-24T14:46:31
|
|
Vulkan: Generate perf warnings on suboptimal paths
Using KHR_debug features, this change creates a
performance-warning-generation macro and employs it in a handful of
locations to provide useful feedback to application developers.
The warnings added in this change are not exhaustive.
Bug: angleproject:3461
Bug: angleproject:4900
Change-Id: Id62435d170d90c5be9c1c5cab2d6779ccb58345e
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2372628
Reviewed-by: Geoff Lang <geofflang@chromium.org>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Commit-Queue: Shahbaz Youssefi <syoussefi@chromium.org>
|
|
|
50442fac
|
2020-08-05T14:15:12
|
|
Vulkan: Fix ImageHelper's move constructor
Bug: angleproject:4913
Change-Id: Ic78a26be4c2f3fa96ef77deffc239dbb7310065e
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2339543
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Tim Van Patten <timvp@google.com>
Commit-Queue: Shahbaz Youssefi <syoussefi@chromium.org>
|
|
|
745e0712
|
2020-03-21T17:46:05
|
|
Vulkan: Enable CPU only buffers for PBOs
Add support for a CPU only buffer for PBOs that serve as
the destination for all host operations like MapBuffer*.
This removes the latency caused by waiting for the in-flight
GPU commands to be complete before handing over the buffer
to the app.
This change removes a ~6ms wait/sleep on the first call to
MapBuffer* in each frame of Manhattan
Bug: angleproject:4339
Tests: angle_end2end_tests --gtest_filter=BufferDataTest*Vulkan
Change-Id: I52016b160af8a670cc30f01c05e48f699521310f
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2116874
Commit-Queue: Mohan Maiya <m.maiya@samsung.com>
Reviewed-by: Tobin Ehlis <tobine@google.com>
|
|
449d9d76
|
2020-03-31T17:27:00
|
|
Vulkan: Refactor garbarge collection related parameter
Use RendererVk instead of VkDevice as a parameter in garbage collection
functions.
Bug: angleproject:2162
Change-Id: Ifd53e05223d6d603402c9b7fcfa82fe1f896458c
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2131882
Commit-Queue: Mohan Maiya <m.maiya@samsung.com>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
|
|
|
a741abb9
|
2020-02-21T16:37:37
|
|
Vulkan: Rename CommandGraphResource to Resource.
Also renames the h and cpp files to ResourceVk (to keep distinct from
other resource.h/cpp files) and renames 'onResourceAccess' to 'retain'.
Cleans up a few remaining mentions of the command graph in comments.
Bug: angleproject:4029
Change-Id: Ifc8e880c8cea3fc48a4aec4730191c88aa35a076
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2065920
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Tim Van Patten <timvp@google.com>
|