Struct CudaDevice 

pub struct CudaDevice {
    pub id: i32,
    pub name: String,
    pub compute_capability: (i32, i32),
    pub total_memory: usize,
    pub free_memory: usize,
    pub multiprocessor_count: i32,
    pub max_threads_per_block: i32,
    pub max_threads_per_mp: i32,
    pub warp_size: i32,
    pub shared_mem_per_block: usize,
    pub default_stream: CudaStream,
    pub memory_pool: Mutex<GpuMemoryPool>,
}

CUDA device handle with real device properties

Fields

id: i32

Device ID

name: String

Device name

compute_capability: (i32, i32)

Compute capability (major, minor)

total_memory: usize

Total memory in bytes

free_memory: usize

Free memory in bytes

multiprocessor_count: i32

Number of multiprocessors

max_threads_per_block: i32

Max threads per block

max_threads_per_mp: i32

Max threads per multiprocessor

warp_size: i32

Warp size

shared_mem_per_block: usize

Shared memory per block

default_stream: CudaStream

Default stream

memory_pool: Mutex<GpuMemoryPool>

Memory pool

Implementations

impl CudaDevice

pub fn init() -> CudaResult<()>

Initialize CUDA runtime

pub fn count() -> CudaResult<i32>

Get number of CUDA devices

pub fn new(device_id: i32) -> CudaResult<Self>

Create device handle for given device ID

pub fn set_current(&self) -> CudaResult<()>

Set current device

pub fn current() -> CudaResult<i32>

Get current device ID

pub fn synchronize() -> CudaResult<()>

Synchronize device - wait for all operations to complete

pub fn get_free_memory(&self) -> CudaResult<usize>

Get current free memory

pub fn properties_string(&self) -> String

Get device properties as string

pub fn has_tensor_cores(&self) -> bool

Check if device supports tensor cores

pub fn supports_fp16(&self) -> bool

Check if device supports FP16

pub fn supports_bf16(&self) -> bool

Check if device supports BF16

pub fn optimal_block_size(&self, shared_mem_per_thread: usize) -> usize

Get optimal block size for a kernel

Trait Implementations

impl Debug for CudaDevice

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

Formats the value using the given formatter.

impl Drop for CudaDevice

fn drop(&mut self)

Executes the destructor for this type.