hal/vulkan: Replace gpu-alloc by gpu-allocator

Author: DeltaEvo

Cargo.lock

@@ -207,15 +207,18 @@ objc = "0.2.5"
 # Vulkan dependencies
 android_system_properties = "0.1.1"
 ash = "0.38"
-gpu-alloc = "0.6"
 gpu-descriptor = "0.3.2"
 
 # DX12 dependencies
-gpu-allocator = { version = "0.27", default-features = false }
 range-alloc = "0.1"
 mach-dxcompiler-rs = { version = "0.1.4", default-features = false } # remember to increase max_shader_model if applicable
 windows-core = { version = "0.58", default-features = false }
 
+# DX12 and Vulkan dependencies
+gpu-allocator = { version = "0.28.0", default-features = false, features = [
+    "hashbrown",
+] }
+
 # Gles dependencies
 khronos-egl = "6"
 glow = "0.16"

Cargo.toml

@@ -91,7 +91,6 @@ vulkan = [
     "dep:arrayvec",
     "dep:ash",
     "dep:bytemuck",
-    "dep:gpu-alloc",
     "dep:gpu-descriptor",
     "dep:hashbrown",
     "dep:libc",
@@ -102,6 +101,7 @@ vulkan = [
     "dep:profiling",
     "dep:smallvec",
     "dep:windows",
+    "gpu-allocator/vulkan",
     "windows/Win32",
 ]
 gles = [
@@ -142,7 +142,7 @@ dx12 = [
     "dep:profiling",
     "dep:range-alloc",
     "dep:windows-core",
-    "dep:gpu-allocator",
+    "gpu-allocator/d3d12",
     "windows/Win32_Graphics_Direct3D_Fxc",
     "windows/Win32_Graphics_Direct3D_Dxc",
     "windows/Win32_Graphics_Direct3D",
@@ -221,9 +221,10 @@ glow = { workspace = true, optional = true }
 ########################
 
 [target.'cfg(not(target_arch = "wasm32"))'.dependencies]
+# Backend: Vulkan and Dx12
+gpu-allocator = { workspace = true, optional = true }
 # Backend: Vulkan
 ash = { workspace = true, optional = true }
-gpu-alloc = { workspace = true, optional = true }
 gpu-descriptor = { workspace = true, optional = true }
 smallvec = { workspace = true, optional = true, features = ["union"] }
 # Backend: GLES
@@ -250,7 +251,6 @@ windows-core = { workspace = true, optional = true }
 # Backend: Dx12
 bit-set = { workspace = true, optional = true }
 range-alloc = { workspace = true, optional = true }
-gpu-allocator = { workspace = true, optional = true, features = ["d3d12"] }
 once_cell = { workspace = true, optional = true }
 # backend: GLES
 glutin_wgl_sys = { workspace = true, optional = true }

wgpu-hal/Cargo.toml

@@ -143,7 +143,7 @@ impl Allocator {
             allocations,
             blocks,
             total_allocated_bytes: upstream.total_allocated_bytes,
-            total_reserved_bytes: upstream.total_reserved_bytes,
+            total_reserved_bytes: upstream.total_capacity_bytes,
         }
     }
 }
@@ -621,37 +621,3 @@ impl<'a> DeviceAllocationContext<'a> {
         Ok(allocation_info)
     }
 }
-
-impl From<gpu_allocator::AllocationError> for crate::DeviceError {
-    fn from(result: gpu_allocator::AllocationError) -> Self {
-        match result {
-            gpu_allocator::AllocationError::OutOfMemory => Self::OutOfMemory,
-            gpu_allocator::AllocationError::FailedToMap(e) => {
-                log::error!("DX12 gpu-allocator: Failed to map: {e}");
-                Self::Lost
-            }
-            gpu_allocator::AllocationError::NoCompatibleMemoryTypeFound => {
-                log::error!("DX12 gpu-allocator: No Compatible Memory Type Found");
-                Self::Lost
-            }
-            gpu_allocator::AllocationError::InvalidAllocationCreateDesc => {
-                log::error!("DX12 gpu-allocator: Invalid Allocation Creation Description");
-                Self::Lost
-            }
-            gpu_allocator::AllocationError::InvalidAllocatorCreateDesc(e) => {
-                log::error!("DX12 gpu-allocator: Invalid Allocator Creation Description: {e}");
-                Self::Lost
-            }
-
-            gpu_allocator::AllocationError::Internal(e) => {
-                log::error!("DX12 gpu-allocator: Internal Error: {e}");
-                Self::Lost
-            }
-            gpu_allocator::AllocationError::BarrierLayoutNeedsDevice10
-            | gpu_allocator::AllocationError::CastableFormatsRequiresEnhancedBarriers
-            | gpu_allocator::AllocationError::CastableFormatsRequiresAtLeastDevice12 => {
-                unreachable!()
-            }
-        }
-    }
-}

wgpu-hal/src/dx12/suballocation.rs

@@ -381,6 +381,41 @@ pub enum DeviceError {
     Unexpected,
 }
 
+#[cfg(any(dx12, vulkan))]
+impl From<gpu_allocator::AllocationError> for DeviceError {
+    fn from(result: gpu_allocator::AllocationError) -> Self {
+        match result {
+            gpu_allocator::AllocationError::OutOfMemory => Self::OutOfMemory,
+            gpu_allocator::AllocationError::FailedToMap(e) => {
+                log::error!("DX12 gpu-allocator: Failed to map: {e}");
+                Self::Lost
+            }
+            gpu_allocator::AllocationError::NoCompatibleMemoryTypeFound => {
+                log::error!("DX12 gpu-allocator: No Compatible Memory Type Found");
+                Self::Lost
+            }
+            gpu_allocator::AllocationError::InvalidAllocationCreateDesc => {
+                log::error!("DX12 gpu-allocator: Invalid Allocation Creation Description");
+                Self::Lost
+            }
+            gpu_allocator::AllocationError::InvalidAllocatorCreateDesc(e) => {
+                log::error!("DX12 gpu-allocator: Invalid Allocator Creation Description: {e}");
+                Self::Lost
+            }
+
+            gpu_allocator::AllocationError::Internal(e) => {
+                log::error!("DX12 gpu-allocator: Internal Error: {e}");
+                Self::Lost
+            }
+            gpu_allocator::AllocationError::BarrierLayoutNeedsDevice10
+            | gpu_allocator::AllocationError::CastableFormatsRequiresEnhancedBarriers
+            | gpu_allocator::AllocationError::CastableFormatsRequiresAtLeastDevice12 => {
+                unreachable!()
+            }
+        }
+    }
+}
+
 #[allow(dead_code)] // may be unused on some platforms
 #[cold]
 fn hal_usage_error<T: fmt::Display>(txt: T) -> ! {

wgpu-hal/src/lib.rs

@@ -2255,87 +2255,42 @@ impl super::Adapter {
             signal_semaphores: Default::default(),
         };
 
-        let mem_allocator = {
-            let limits = self.phd_capabilities.properties.limits;
-
-            // Note: the parameters here are not set in stone nor where they picked with
-            // strong confidence.
-            // `final_free_list_chunk` should be bigger than starting_free_list_chunk if
-            // we want the behavior of starting with smaller block sizes and using larger
-            // ones only after we observe that the small ones aren't enough, which I think
-            // is a good "I don't know what the workload is going to be like" approach.
-            //
-            // For reference, `VMA`, and `gpu_allocator` both start with 256 MB blocks
-            // (then VMA doubles the block size each time it needs a new block).
-            // At some point it would be good to experiment with real workloads
-            //
-            // TODO(#5925): The plan is to switch the Vulkan backend from `gpu_alloc` to
-            // `gpu_allocator` which has a different (simpler) set of configuration options.
-            //
-            // TODO: These parameters should take hardware capabilities into account.
-            let mb = 1024 * 1024;
-            let perf_cfg = gpu_alloc::Config {
-                starting_free_list_chunk: 128 * mb,
-                final_free_list_chunk: 512 * mb,
-                minimal_buddy_size: 1,
-                initial_buddy_dedicated_size: 8 * mb,
-                dedicated_threshold: 32 * mb,
-                preferred_dedicated_threshold: mb,
-                transient_dedicated_threshold: 128 * mb,
-            };
-            let mem_usage_cfg = gpu_alloc::Config {
-                starting_free_list_chunk: 8 * mb,
-                final_free_list_chunk: 64 * mb,
-                minimal_buddy_size: 1,
-                initial_buddy_dedicated_size: 8 * mb,
-                dedicated_threshold: 8 * mb,
-                preferred_dedicated_threshold: mb,
-                transient_dedicated_threshold: 16 * mb,
-            };
-            let config = match memory_hints {
-                wgt::MemoryHints::Performance => perf_cfg,
-                wgt::MemoryHints::MemoryUsage => mem_usage_cfg,
-                wgt::MemoryHints::Manual {
-                    suballocated_device_memory_block_size,
-                } => gpu_alloc::Config {
-                    starting_free_list_chunk: suballocated_device_memory_block_size.start,
-                    final_free_list_chunk: suballocated_device_memory_block_size.end,
-                    initial_buddy_dedicated_size: suballocated_device_memory_block_size.start,
-                    ..perf_cfg
-                },
-            };
-
-            let max_memory_allocation_size =
-                if let Some(maintenance_3) = self.phd_capabilities.maintenance_3 {
-                    maintenance_3.max_memory_allocation_size
-                } else {
-                    u64::MAX
-                };
-            let properties = gpu_alloc::DeviceProperties {
-                max_memory_allocation_count: limits.max_memory_allocation_count,
-                max_memory_allocation_size,
-                non_coherent_atom_size: limits.non_coherent_atom_size,
-                memory_types: memory_types
-                    .iter()
-                    .map(|memory_type| gpu_alloc::MemoryType {
-                        props: gpu_alloc::MemoryPropertyFlags::from_bits_truncate(
-                            memory_type.property_flags.as_raw() as u8,
-                        ),
-                        heap: memory_type.heap_index,
-                    })
-                    .collect(),
-                memory_heaps: mem_properties
-                    .memory_heaps_as_slice()
-                    .iter()
-                    .map(|&memory_heap| gpu_alloc::MemoryHeap {
-                        size: memory_heap.size,
-                    })
-                    .collect(),
-                buffer_device_address: enabled_extensions
-                    .contains(&khr::buffer_device_address::NAME),
-            };
-            gpu_alloc::GpuAllocator::new(config, properties)
+        // TODO: the allocator's configuration should take hardware capability into
+        // account.
+        const MB: u64 = 1024 * 1024;
+        let allocation_sizes = match memory_hints {
+            wgt::MemoryHints::Performance => gpu_allocator::AllocationSizes::new(128 * MB, 64 * MB)
+                .with_max_device_memblock_size(256 * MB)
+                .with_max_host_memblock_size(128 * MB),
+            wgt::MemoryHints::MemoryUsage => gpu_allocator::AllocationSizes::new(8 * MB, 4 * MB)
+                .with_max_device_memblock_size(64 * MB)
+                .with_max_host_memblock_size(32 * MB),
+            wgt::MemoryHints::Manual {
+                suballocated_device_memory_block_size,
+            } => {
+                // TODO: Would it be useful to expose the host size in memory hints
+                // instead of always using half of the device size?
+                let device_size = suballocated_device_memory_block_size;
+                let host_size = device_size.start / 2..device_size.end / 2;
+
+                gpu_allocator::AllocationSizes::new(device_size.start, host_size.start)
+                    .with_max_device_memblock_size(device_size.end)
+                    .with_max_host_memblock_size(host_size.end)
+            }
         };
+
+        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 desc_allocator = gpu_descriptor::DescriptorAllocator::new(
             if let Some(di) = self.phd_capabilities.descriptor_indexing {
                 di.max_update_after_bind_descriptors_in_all_pools