diff --git a/wgpu-core/src/resource.rs b/wgpu-core/src/resource.rs
index 9c2252e665..6418053f1b 100644
--- a/wgpu-core/src/resource.rs
+++ b/wgpu-core/src/resource.rs
@@ -8,7 +8,10 @@ use crate::{
     },
     global::Global,
     hal_api::HalApi,
-    id::{AdapterId, BufferId, CommandEncoderId, DeviceId, SurfaceId, TextureId, TextureViewId},
+    id::{
+        AdapterId, BufferId, CommandEncoderId, DeviceId, QueueId, SurfaceId, TextureId,
+        TextureViewId,
+    },
     init_tracker::{BufferInitTracker, TextureInitTracker},
     lock::{rank, Mutex, RwLock},
     resource_log,
@@ -1388,6 +1391,21 @@ impl Global {
             hal_command_encoder_callback(None)
         }
     }
+
+    /// # Safety
+    ///
+    /// - The raw queue handle must not be manually destroyed
+    pub unsafe fn queue_as_hal<A: HalApi, F, R>(&self, id: QueueId, hal_queue_callback: F) -> R
+    where
+        F: FnOnce(Option<&A::Queue>) -> R,
+    {
+        profiling::scope!("Queue::as_hal");
+
+        let queue = self.hub.queues.get(id);
+        let hal_queue = queue.raw().as_any().downcast_ref();
+
+        hal_queue_callback(hal_queue)
+    }
 }
 
 /// A texture that has been marked as destroyed and is staged for actual deletion soon.
diff --git a/wgpu-hal/src/vulkan/adapter.rs b/wgpu-hal/src/vulkan/adapter.rs
index fd190fc34a..61020c2a34 100644
--- a/wgpu-hal/src/vulkan/adapter.rs
+++ b/wgpu-hal/src/vulkan/adapter.rs
@@ -1982,6 +1982,7 @@ impl super::Adapter {
             device: Arc::clone(&shared),
             family_index,
             relay_semaphores: Mutex::new(relay_semaphores),
+            signal_semaphores: Mutex::new((Vec::new(), Vec::new())),
         };
 
         let mem_allocator = {
diff --git a/wgpu-hal/src/vulkan/mod.rs b/wgpu-hal/src/vulkan/mod.rs
index 31a893e56e..b0d27a9a36 100644
--- a/wgpu-hal/src/vulkan/mod.rs
+++ b/wgpu-hal/src/vulkan/mod.rs
@@ -36,6 +36,7 @@ use std::{
     ffi::{CStr, CString},
     fmt, mem,
     num::NonZeroU32,
+    ops::DerefMut,
     sync::Arc,
 };
 
@@ -765,6 +766,7 @@ pub struct Queue {
     device: Arc<DeviceShared>,
     family_index: u32,
     relay_semaphores: Mutex<RelaySemaphores>,
+    signal_semaphores: Mutex<(Vec<vk::Semaphore>, Vec<u64>)>,
 }
 
 impl Drop for Queue {
@@ -1216,6 +1218,15 @@ impl crate::Queue for Queue {
             signal_values.push(!0);
         }
 
+        let mut guards = self.signal_semaphores.lock();
+        let (ref mut pending_signal_semaphores, ref mut pending_signal_semaphore_values) =
+            guards.deref_mut();
+        assert!(pending_signal_semaphores.len() == pending_signal_semaphore_values.len());
+        if !pending_signal_semaphores.is_empty() {
+            signal_semaphores.append(pending_signal_semaphores);
+            signal_values.append(pending_signal_semaphore_values);
+        }
+
         // In order for submissions to be strictly ordered, we encode a dependency between each submission
         // using a pair of semaphores. This adds a wait if it is needed, and signals the next semaphore.
         let semaphore_state = self.relay_semaphores.lock().advance(&self.device)?;
@@ -1344,6 +1355,19 @@ impl crate::Queue for Queue {
     }
 }
 
+impl Queue {
+    pub fn raw_device(&self) -> &ash::Device {
+        &self.device.raw
+    }
+
+    pub fn add_signal_semaphore(&self, semaphore: vk::Semaphore, semaphore_value: Option<u64>) {
+        let mut guards = self.signal_semaphores.lock();
+        let (ref mut semaphores, ref mut semaphore_values) = guards.deref_mut();
+        semaphores.push(semaphore);
+        semaphore_values.push(semaphore_value.unwrap_or(!0));
+    }
+}
+
 /// Maps
 ///
 /// - VK_ERROR_OUT_OF_HOST_MEMORY