-
Notifications
You must be signed in to change notification settings - Fork 26.7k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[Impeller] Vulkan: Clearing signaled fences and calling dtors stalls fence waiter. #129737
Comments
I guess you can't say "fix" and issue too close together. |
…pport memoryless. (#43349) For devices that don't support memoryless textures, the fullscreen MSAA texture adds a substantial amount of memory thrashing when it is allocated and deallocated. For these devices, lets cache the MSAA texture in the swapchain image. flutter/flutter#129737
We have fixed the issue of the dtors being slow which effectively fixes the stall. We should figure out if there is anything to do w.r.t tossing the dtors onto the pool and off the waiter thread so we can never stall on a user workload. |
…fer flags. (#43313) Following some thresholds from ANGLE, lets try allocating fewer of our resources into dedicated memory to see if that improves allocator performance. flutter/flutter#129737
Some Android devices do not support the memory type eLazilyAllocated, which we use for MSAA and stencil textures. These textures were falling back to device local in dedicated allocations, which are expensive to both allocate and free. The dedicated allocation is implied by asking for eLazilyAllocated Instead, perform a check for support for this memory type. Never request dedicated allocations (at least not until we have a compelling use case) This should fix flutter/flutter#129737 flutter/flutter#129784
…pport memoryless. (flutter#43349) For devices that don't support memoryless textures, the fullscreen MSAA texture adds a substantial amount of memory thrashing when it is allocated and deallocated. For these devices, lets cache the MSAA texture in the swapchain image. flutter/flutter#129737
…fer flags. (flutter#43313) Following some thresholds from ANGLE, lets try allocating fewer of our resources into dedicated memory to see if that improves allocator performance. flutter/flutter#129737
This thread has been automatically locked since there has not been any recent activity after it was closed. If you are still experiencing a similar issue, please open a new bug, including the output of |
This will delay all further deallocations. On this run on CI, the waiter was stalled for more than 5 frames. That means the tracked resources in the command buffer were not collected during this time. This increases peak memory usage and probably also puts pressure on the allocator as it cannot reuse memory and must allocate newer arenas. The trace is added here.
The text was updated successfully, but these errors were encountered: