Vulkan backend crash with WGSL dot function

[hanawatson]

Description
wgpu with Vulkan backend crashes with a parent device is lost validation error when given a WGSL shader that uses the dot function.

Repro steps
Run this WGSL code (it seems to work for all alternative random values of vectors that I have tried) through wgpu with Vulkan backend:

struct outputBuffer {
	output: i32,
}

@group(0) @binding(0)
var<storage, read_write> out: outputBuffer;

@compute @workgroup_size(1)
fn main() {
	var var0 = dot(vec2<i32>(1, 2), vec2<i32>(1, 2));
	out.output = 0;
}

Expected vs observed behavior
Expecting normal compilation/running behavior, and an output of 0 for the out buffer. Instead, wgpu seems to crash:

wgpu error: Validation Error
        
        Caused by:
            In Device::create_compute_pipeline
            parent device is lost

Extra materials
Setting RUST_LOG=trace gets a lot of information about my GPU etc., followed by:

[2022-05-25T18:59:33Z DEBUG naga::front] Resolving [1] = GlobalVariable([1]) : Value(Pointer { base: [2], space: Storage { access: LOAD | STORE } })
        [2022-05-25T18:59:33Z DEBUG naga::front] Resolving [2] = Constant([1]) : Value(Scalar { kind: Sint, width: 4 })
        [2022-05-25T18:59:33Z DEBUG naga::front] Resolving [3] = Constant([2]) : Value(Scalar { kind: Sint, width: 4 })
        [2022-05-25T18:59:33Z DEBUG naga::front] Resolving [4] = Compose { ty: [3], components: [[2], [3]] } : Handle([3])
        [2022-05-25T18:59:33Z DEBUG naga::front] Resolving [5] = Constant([1]) : Value(Scalar { kind: Sint, width: 4 })
        [2022-05-25T18:59:33Z DEBUG naga::front] Resolving [6] = Constant([2]) : Value(Scalar { kind: Sint, width: 4 })
        [2022-05-25T18:59:33Z DEBUG naga::front] Resolving [7] = Compose { ty: [3], components: [[5], [6]] } : Handle([3])
        [2022-05-25T18:59:33Z DEBUG naga::front] Resolving [8] = Math { fun: Dot, arg: [4], arg1: Some([7]), arg2: None, arg3: None } : Value(Scalar { kind: Sint, width: 4 })
        [2022-05-25T18:59:33Z DEBUG naga::front] Resolving [9] = LocalVariable([1]) : Value(Pointer { base: [1], space: Function })
        [2022-05-25T18:59:33Z DEBUG naga::front] Resolving [10] = AccessIndex { base: [1], index: 0 } : Value(Pointer { base: [1], space: Storage { access: LOAD | STORE } })
        [2022-05-25T18:59:33Z DEBUG naga::front] Resolving [11] = Constant([3]) : Value(Scalar { kind: Sint, width: 4 })
        [2022-05-25T18:59:33Z DEBUG naga::valid::interface] var GlobalVariable { name: Some("checksum"), space: Storage { access: LOAD | STORE }, binding: Some(ResourceBinding { group: 0, binding: 0 }), ty: [2], init: None }
        [2022-05-25T18:59:33Z DEBUG naga::valid::function] var LocalVariable { name: Some("var0"), ty: [1], init: None }
        [2022-05-25T18:59:33Z WARN  wgpu_hal::vulkan] Unrecognized device error ERROR_UNKNOWN
        [2022-05-25T18:59:33Z ERROR wgpu::backend::direct] Handling wgpu errors as fatal by default

and enabling Vulkan validation layers gives this:

UNASSIGNED-CoreValidation-Shader-InconsistentSpirv(ERROR / SPEC): msgNum: 7060244 - Validation Error: [ UNASSIGNED-CoreValidation-Shader-InconsistentSpirv ] Object 0: handle = 0x55fb1f2e70f0, type = VK_OBJECT_TYPE_DEVICE; | MessageID = 0x6bbb14 | SPIR-V module not valid: ConstantNull cannot appear in a function declaration
          %26 = OpConstantNull %int
        
            Objects: 1
                [0] 0x55fb1f2e70f0, type: 3, name: NULL
        VUID-vkDestroyDevice-device-00378(ERROR / SPEC): msgNum: 1901072314 - Validation Error: [ VUID-vkDestroyDevice-device-00378 ] Object 0: handle = 0x55fb1f2e70f0, type = VK_OBJECT_TYPE_DEVICE; Object 1: handle = 0xee647e0000000009, type = VK_OBJECT_TYPE_SHADER_MODULE; | MessageID = 0x71500fba | OBJ ERROR : For VkDevice 0x55fb1f2e70f0[], VkShaderModule 0xee647e0000000009[] has not been destroyed. The Vulkan spec states: All child objects created on device must have been destroyed prior to destroying device (https://vulkan.lunarg.com/doc/view/1.3.211.0/linux/1.3-extensions/vkspec.html#VUID-vkDestroyDevice-device-00378)
            Objects: 2
                [0] 0x55fb1f2e70f0, type: 3, name: NULL
                [1] 0xee647e0000000009, type: 15, name: NULL

Platform
cargo 1.60.0, wgpu v0.12.0, Mesa Intel(R) UHD Graphics 620 (KBL GT2) (renderer version 21.3.8), Vulkan backend, Kali Linux
(Was planning on getting a little more information using wgpu-info, but every time I try to run it panics with a No such file error - unsure if this is relevant)

[cwfitzgerald]

Is this against master or against released 0.12?

[hanawatson]

Is this against master or against released 0.12?

Against released 0.12, I think (the crate version is v0.12.0, so I'm assuming these match up!).

[jinleili]

Verified on master, cannot reproduce this issue

[cwfitzgerald]

Thank you so much for this!