Port Vulkan backend to in-tree suballocator

Port the Vulkan backend off gpu-allocator onto the new in-repo
wgpu-block-pool / wgpu-offset-allocator crates via a new
vulkan/suballocation.rs (VMA-style memory-type selection, per-heap block
sizing, dedicated-allocation heuristics, budget gating, persistent
mapping). Vulkan no longer references lib.rs's shared AllocationSizes; it
uses a local BlockSizePolicy instead. Also hardens acceleration-structure
buffer alignment and fixes device-memory leaks on error paths.

The dx12 backend still uses gpu-allocator, so the shared gpu_allocator
glue in lib.rs (gated on any(dx12, vulkan)) is kept; the gpu-allocator
crate remains a wgpu-hal dependency (its vulkan feature is dropped, its
d3d12 feature retained). Removing that glue is a follow-up once dx12 is
ported too.

Author: cwfitzgerald

Cargo.lock

@@ -94,6 +94,12 @@ vulkan = [
     "dep:arrayvec",
     "dep:ash",
     "dep:bytemuck",
+    # The vulkan backend no longer uses gpu-allocator, but the shared
+    # `gpu_allocator`-based glue in `lib.rs` is still gated on `any(dx12, vulkan)`
+    # (the dx12 backend continues to use it). Keep the crate available - without its
+    # `vulkan` feature - so `lib.rs` compiles in vulkan-only builds. Removing that glue
+    # (and this dependency) is a follow-up once dx12 is ported too.
+    "dep:gpu-allocator",
     "dep:libc",
     "dep:libloading",
     "dep:ordered-float",
@@ -103,7 +109,8 @@ vulkan = [
     "dep:smallvec",
     "dep:windows",
     "dep:windows-core",
-    "gpu-allocator/vulkan",
+    "dep:wgpu-block-pool",
+    "dep:wgpu-offset-allocator",
     "windows/Win32",
     "windows/Win32_Devices_Display",
     "windows/Win32_Foundation",
@@ -211,6 +218,9 @@ naga.workspace = true
 wgpu-naga-bridge.workspace = true
 naga-types = { workspace = true }
 wgpu-types = { workspace = true, default-features = false }
+# Backend: Vulkan (in-repo suballocator)
+wgpu-block-pool = { workspace = true, optional = true }
+wgpu-offset-allocator = { workspace = true, optional = true }
 
 # Dependencies in the lib and empty backend
 bitflags.workspace = true

wgpu-hal/Cargo.toml

@@ -4,7 +4,7 @@ use core::{ffi::CStr, marker::PhantomData};
 use ash::{ext, google, khr, vk};
 use parking_lot::Mutex;
 
-use crate::{vulkan::semaphore_list::SemaphoreList, AllocationSizes};
+use crate::vulkan::semaphore_list::SemaphoreList;
 
 use super::semaphore_list::SemaphoreListMode;
 
@@ -2910,6 +2910,7 @@ impl super::Adapter {
             drop_guard,
             instance: Arc::clone(&self.instance),
             physical_device: self.raw,
+            device_api_version: self.phd_capabilities.device_api_version,
             enabled_extensions: enabled_extensions.into(),
             extension_fns: super::DeviceExtensionFunctions {
                 debug_utils: debug_utils_fn,
@@ -2948,19 +2949,18 @@ impl super::Adapter {
             wait_semaphores: Mutex::new(SemaphoreList::new(SemaphoreListMode::Wait)),
         };
 
-        let allocation_sizes = AllocationSizes::from_memory_hints(memory_hints).into();
-
+        let memory_budget_supported = enabled_extensions.contains(&ext::memory_budget::NAME);
         let buffer_device_address = enabled_extensions.contains(&khr::buffer_device_address::NAME);
 
-        let mem_allocator =
-            gpu_allocator::vulkan::Allocator::new(&gpu_allocator::vulkan::AllocatorCreateDesc {
-                instance: self.instance.raw.clone(),
-                device: shared.raw.clone(),
-                physical_device: self.raw,
-                debug_settings: Default::default(),
-                buffer_device_address,
-                allocation_sizes,
-            })?;
+        let mem_allocator = super::suballocation::Allocator::new(
+            &mem_properties,
+            &self.phd_capabilities.properties.limits,
+            valid_ash_memory_types,
+            memory_hints,
+            self.instance.memory_budget_thresholds,
+            memory_budget_supported,
+            buffer_device_address,
+        );
 
         let desc_allocator = super::descriptor::DescriptorAllocator::new(
             if let Some(di) = self.phd_capabilities.descriptor_indexing {
@@ -2974,7 +2974,6 @@ impl super::Adapter {
             shared,
             mem_allocator: Mutex::new(mem_allocator),
             desc_allocator: Mutex::new(desc_allocator),
-            valid_ash_memory_types,
             naga_options,
             #[cfg(feature = "renderdoc")]
             render_doc: Default::default(),