SwiftShader: panic in wgpu-hal: Invariant violated when requesting device

[mandryskowski]

Description
Calling request_device results in a panic inside wgpu-hal when using SwiftShader.

Repro steps
I have created a minimal reproducible example here:
https://github.com/mandryskowski/wgpu-swiftshader-feature-crash/blob/master/src/main.rs
See this Github Actions run https://github.com/mandryskowski/wgpu-swiftshader-feature-crash/actions/runs/23510896074/job/68431261257

let (device, _queue) = adapter.request_device(
	&DeviceDescriptor {
		required_features: Features::empty(),
		..Default::default()
	}
).await.unwrap();

Actual behavior

Found Adapter: AdapterInfo { name: "SwiftShader Device (LLVM 10.0.0)", vendor: 6880, device: 49374, device_type: Cpu, device_pci_bus_id: "", driver: "SwiftShader driver", driver_info: "", backend: Vulkan, subgroup_min_size: 4, subgroup_max_size: 4, transient_saves_memory: false, limit_bucket: None }
[2026-03-24T18:02:07Z ERROR wgpu_hal::vulkan::instance] GENERAL [Loader Message (0x0)]
    	terminator_CreateDevice: Failed in ICD /home/runner/work/wgpu-swiftshader-feature-crash/wgpu-swiftshader-feature-crash/swiftshader/build/Linux/./libvk_swiftshader.so vkCreateDevice call
[2026-03-24T18:02:07Z ERROR wgpu_hal::vulkan::instance] 	objects: (type: INSTANCE, hndl: 0x55ad69579620, name: ?)
[2026-03-24T18:02:07Z ERROR wgpu_hal::vulkan::instance] GENERAL [Loader Message (0x0)]
    	vkCreateDevice:  Failed to create device chain.
[2026-03-24T18:02:07Z ERROR wgpu_hal::vulkan::instance] 	objects: (type: INSTANCE, hndl: 0x55ad69579620, name: ?)

thread 'main' (10113) panicked at wgpu/wgpu-hal/src/lib.rs:489:5:
wgpu-hal invariant was violated (usage error): Requested feature is not available on this device
stack backtrace:
   0: __rustc::rust_begin_unwind
             at /rustc/4a4ef493e3a1488c6e321570238084b38948f6db/library/std/src/panicking.rs:689:5
   1: core::panicking::panic_fmt
             at /rustc/4a4ef493e3a1488c6e321570238084b38948f6db/library/core/src/panicking.rs:80:14
   2: wgpu_hal::hal_usage_error
             at ./wgpu/wgpu-hal/src/lib.rs:489:5
   3: wgpu_hal::vulkan::adapter::<impl wgpu_hal::vulkan::Adapter>::open_with_callback::map_err
             at ./wgpu/wgpu-hal/src/vulkan/adapter.rs:2865:21
   4: core::ops::function::FnOnce::call_once
             at /rustc/4a4ef493e3a1488c6e321570238084b38948f6db/library/core/src/ops/function.rs:250:5
   5: core::result::Result<T,E>::map_err
             at /rustc/4a4ef493e3a1488c6e321570238084b38948f6db/library/core/src/result.rs:968:27
   6: wgpu_hal::vulkan::adapter::<impl wgpu_hal::vulkan::Adapter>::open_with_callback
             at ./wgpu/wgpu-hal/src/vulkan/adapter.rs:2857:22
   7: wgpu_hal::vulkan::adapter::<impl wgpu_hal::Adapter for wgpu_hal::vulkan::Adapter>::open
             at ./wgpu/wgpu-hal/src/vulkan/adapter.rs:2895:23
   8: <A as wgpu_hal::dynamic::adapter::DynAdapter>::open
             at ./wgpu/wgpu-hal/src/dynamic/adapter.rs:52:18
   9: wgpu_core::instance::Adapter::create_device_and_queue
             at ./wgpu/wgpu-core/src/instance.rs:871:30
  10: wgpu_core::instance::<impl wgpu_core::global::Global>::adapter_request_device
             at ./wgpu/wgpu-core/src/instance.rs:1202:39
  11: <wgpu::backend::wgpu_core::CoreAdapter as wgpu::dispatch::AdapterInterface>::request_device
             at ./wgpu/wgpu/src/backend/wgpu_core.rs:936:34
  12: wgpu::api::adapter::Adapter::request_device
             at ./wgpu/wgpu/src/api/adapter.rs:62:33
  13: wgpu_subgroup_panic_repro::run::{{closure}}
             at ./src/main.rs:34:10
  14: pollster::block_on
             at /home/runner/.cargo/registry/src/index.crates.io-1949cf8c6b5b557f/pollster-0.3.0/src/lib.rs:128:28
  15: wgpu_subgroup_panic_repro::main
             at ./src/main.rs:5:5
  16: core::ops::function::FnOnce::call_once
             at /rustc/4a4ef493e3a1488c6e321570238084b38948f6db/library/core/src/ops/function.rs:250:5
note: Some details are omitted, run with `RUST_BACKTRACE=full` for a verbose backtrace.
Error: Process completed with exit code 101.

Platform
swiftshader 89556131bf9d48af3c5c9fbb9a3322e706da89a3 (recent)
wgpu 0226d561c03e99d6805afb989e632198f5da5d08 (recent)
Ubuntu 24.04.3 LTS

[mandryskowski]

I bisected this and it seems that 7609ed99e59e2486238caa0a4f7fe80d4c132ef3 broke this

[andyleiserson]

I saw the same wgpu-hal invariant was violated (usage error): Requested feature is not available on this device running binding_array tests on MoltenVK.

Claude claimed "MoltenVK has a bug where chaining ANY VK 1.2 promoted struct (like descriptor_indexing, which is required for this test) triggers validation of ALL VK 1.2 features including vulkanMemoryModel.", but on further prodding it couldn't substantiate that claim. So not sure how useful this information is, but posting it on the offchance it helps narrow down some problem in how Vulkan device capabilities are managed.

[teoxoy]

We are getting VK_ERROR_FEATURE_NOT_PRESENT from vkCreateDevice.

wgpu/wgpu-hal/src/vulkan/adapter.rs

Line 2864 in 73600f1

vk::Result::ERROR_EXTENSION_NOT_PRESENT | vk::Result::ERROR_FEATURE_NOT_PRESENT => {

So it seems like we are somehow trying to create a device with a feature enabled that it doesn't support.

I bisected this and it seems that 7609ed99e59e2486238caa0a4f7fe80d4c132ef3 broke this

This is likely the cause (added in #8883):

wgpu/wgpu-hal/src/vulkan/adapter.rs

Line 623 in 83923e0

.shader_draw_parameters(true),

It should only be enabled if SHADER_DRAW_INDEX was requested.

cc @inner-daemons