archive-gh-me/wgpu
1
0
Fork 0
mirror of https://github.com/gfx-rs/wgpu.git synced 2025-12-08 21:26:17 +00:00
Code Issues Projects Releases Wiki Activity

Skip wgpu_gpu::oom::query_set_oom_test on Vulkan.

Browse Source 
Vulkan's `VK_EXT_memory_budget` extension doesn't cover query sets, so
`wgpu_hal::vulkan::Device::error_if_would_oom_on_resource_allocation`'s
tactic doesn't work to anticipate OOM errors from the Vulkan driver
when creating query sets.

Without these checks, `wgpu_gpu::oom::query_set_oom_test` actually
consumes a significant amount of memory and attracts the attention of
the system's OOM killer. While it does get killed, the situation tends
to adversely affect the stability of the rest of the system, and thus
is not a friendly thing for a test run to do (#7817).

This change can be reverted once a suitable accounting method for
query sets has been implemented on Vulkan.
This commit is contained in:
Jim Blandy 2025-06-17 14:18:25 -07:00
parent 3a5d0f2747
commit 3f96ba2980
1 changed files with 5 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
tests/tests/wgpu-gpu/oom.rs
View File

@ -13,6 +13,9 @@ use wgpu_test::{gpu_test, FailureCase, GpuTestConfiguration, TestParameters};
/// Backends for which OOM detection is implemented
const OOM_DETECTION_IMPL: Backends = Backends::DX12.union(Backends::VULKAN);
/// Backends for which query set OOM detection is implemented
const QUERY_SET_OOM_DETECTION_IMPL: Backends = Backends::DX12;
// All tests skip llvmpipe.
// Even though llvmpipe supports VK_EXT_memory_budget it's happy to continue creating resources until
// the process crashes with SIGABRT "memory allocation of X bytes failed" or the test times out.
@ -125,7 +128,8 @@ static MAPPING_BUFFER_OOM_TEST: GpuTestConfiguration = GpuTestConfiguration::new
static QUERY_SET_OOM_TEST: GpuTestConfiguration = GpuTestConfiguration::new()
.parameters(
TestParameters::default()
.skip(FailureCase::backend(!OOM_DETECTION_IMPL))
// Vulkan: https://github.com/gfx-rs/wgpu/issues/7817
.skip(FailureCase::backend(!QUERY_SET_OOM_DETECTION_IMPL))
// see comment at the top of the file
.skip(FailureCase::backend_adapter(Backends::VULKAN, "llvmpipe")),
)