Struct DeviceCapabilities 

pub struct DeviceCapabilities {
    pub preferred_surface_format: TextureFormat,
    pub preferred_render_target_format: TextureFormat,
    pub supported_surface_formats: Vec<TextureFormat>,
    pub supported_render_target_formats: Vec<TextureFormat>,
    pub has_zero_copy_storage_readback: bool,
    pub buffer_resize_cost: BufferResizeCost,
    pub buffer_page_size: u64,
    pub buffer_decommit_supported: bool,
    pub host_sidecar_on_submit_worker: bool,
    pub split_compute_partitions_on_barrier_cost: bool,
    pub fuse_upload_with_compute_partitions: bool,
}

Device capabilities and format preferences.

Use this to query the optimal formats and limits for your use case.

Fields

preferred_surface_format: TextureFormat

Preferred format for window surfaces (swapchains). For windowed apps, use this for RenderPipelineDesc::target_format.

preferred_render_target_format: TextureFormat

Preferred format for off-screen render targets. For headless rendering (video encoding, CPU readback), use this format.

supported_surface_formats: Vec<TextureFormat>

Formats supported for window surfaces.

supported_render_target_formats: Vec<TextureFormat>

Formats supported for render targets.

has_zero_copy_storage_readback: bool

Whether crate::types::BufferFlags::CPU_READABLE scattered buffers can be read from CPU without a GPU copy.

true on Vulkan (HOST_VISIBLE storage) and Metal (Shared storage). false on Direct3D 12 (requires GPU copy to a READBACK heap).

buffer_resize_cost: BufferResizeCost

How costly in-place buffer resize (resize_to) is on this device.

buffer_page_size: u64

Sparse / tile page size when applicable; informational for aligning resize hints.

buffer_decommit_supported: bool

Whether hint_unused_above on backing buffer allocations can return physical memory to the system.

host_sidecar_on_submit_worker: bool

Whether crate::Scheme::defer_host_write / host-observed waits are applied on the submission worker before GPU execute (DX12, Vulkan, and Metal).

When false, callers must keep synchronous host writes and render-thread reuse gates.

split_compute_partitions_on_barrier_cost: bool

Whether large pure-compute partitions may be subdivided at their heaviest barrier boundary to expose GPU-pipeline overlap between submissions.

Enabled on Vulkan/DX12. Disabled on Metal: cross-CB MTLSharedEvent waits serialize consecutive partitions and dominate any overlap gains.

fuse_upload_with_compute_partitions: bool

Whether the fresh Scheme submit path may fuse an upload-only partition with the immediately following compute partition into one command buffer.

Enabled on Metal so blit uploads and compute dispatches share one MTLCommandBuffer (blit encoder → compute encoder). Disabled elsewhere: upload and compute partitions remain separate submissions.

Trait Implementations

impl Clone for DeviceCapabilities

fn clone(&self) -> DeviceCapabilities

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for DeviceCapabilities

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for DeviceCapabilities

fn default() -> Self

Returns the “default value” for a type.